Instead put them on the most recent valid packet on the page.
Also bullet-proof the offset checking to the "use the current
 position when seeking" code added in 6d61f3f1.
The previous code relied on the file not changing out from under
 us, which we shouldn't do.

This is no longer guaranteed after the change in e7661191, so test
 for it explicitly.
Thanks to Greg Maxwell for the test case.

This case can basically only get hit if there's another stream
 multiplexed in with ours, but this avoids re-scanning the same
 data multiple times when it happens.

In most cases, this is a minor speed-up at best, but in some it can
 be a big win.
Also, when forcing a true bisection instead of a secant-step,
 don't add the negative bias to the result.
This can significantly improve the worst-case.

- Attempt to re-use connections when we've already received enough
   data to do so immediately.
- Make sure when seeking near the end, if the current chunk size is
   such that the _next_ chunk will be half the normal size or less,
   we just ask for the rest of the resource.

With these two changes, a normal open of a single-chain Opus-only
 file requires exactly two HTTP requests.

- Also use the response buffer as a dummy buffer when skipping
   data.
  This will avoid helgrind errors for multiple writes from
   different threads without locking (should someone be reading
   multiple streams from different threads).
  It's also better for SMP cache contention.

This can be quite expensive with the http backend, especially if it
 causes us to pass a chunk threshold and issue a new request.
It also lets us error out more quickly if the underlying stream
 data changes.

We apply a positive bias when the previous bisection point was
 inside the current link, causing us to scan forward a bit instead
 of seeking to a new location.
This knocks up to 18% off the number of seeks required to open very
 large files with lots of links.

- The DIGIT character sets shouldn't need to list "0" twice.
- Avoid a lookup for the port number in getaddrinfo().
- Resolve the OPUS_SET_GAIN TODO (by refusing to implement a fallback).
- A few more minor things.

This changes op_read_func to
a) Take a single byte count to read instead of an "item" count
    (which the http backend couldn't properly support anyway).
b) Use integers for buffer sizes to avoid having to worry about
    sign differences and whether size_t is larger or smaller than
    opus_int64, etc.
c) Return an explicit error code (instead of using errno like
    fread).
   We had already eliminated the use of errno, but we did it by
    treating read errors and EOF identically in all cases.
   This was preventing us from reporting SSL truncation attacks
    from the https backend.
   The https backend now properly reports such errors.

This commit also fixes a bug introduced in 9b57b0c2, where we
 accidentally started passing absolute offsets to the _boundary
 parameter of op_get_next_page() instead of relative offsets.
We now use absolute offsets in all places, as it is the simpler
 choice.
This matters now, because the error reported when encountering EOF
 before hitting the _boundary is no longer suppressed (but instead
 reported as OP_EBADLINK).

Finally, it removes the op_page_seek() function.
Except for the time needed to decode forward after seeking, this
 function was identical in performance to op_pcm_seek(), and Opus
 requires decoding 80 ms of data after seek anyway, so the relative
 benefit is much smaller than with Vorbis.
A survey of open-source code using libvorbisfile showed that the
 only usages of ov_page_seek() in the wild were calling it to seek
 to the start of the stream, for which op_pcm_seek() already has a
 special case that makes it just as fast.

The documentation was also updated to describe all of these chanes.

This is an incompatible API change.

Right now we have no way to add any more parameters beyond a set of
 basic binary flags.
This unifies op_url_stream_create() and
 op_url_stream_create_with_proxy() into a single function that
 takes a variable-length list of arguments, which can be extended
 in the future to include more options of any type.

This is an incompatible API change.

We were computing the return value correctly, but then not
 returning it.

- Increase the maximum response header buffer size up to ~32 kB.
  This also moves it into a heap-allocated buffer instead of the
   stack, as 32 kB is really too much for the stack.
- Treat LF as CR LF when parsing headers.
  This is necessary when parsing the load-balancer response in
   front of <http://lazaradio.hu:8100/bermuda.opus>.
  The response returned by that server is invalid in lots of ways,
   but with these two changes we can read it.
- In addition, we now peek ahead at a large chunk of data when
   reading the response instead of reading 2 to 4 bytes at a time.
  This allows a typical response to be read with two syscalls
   (one peek, one read) instead of several hundred.
- Stop trying to read more data when the connection is closed.

Thanks to Jean-Marc Valin for providing documentation feedback.
This also changes op_tags() to allow the tags to be queried on a
 partially-open file (because they've already been read and are
 available), and fixes a leak when a partially open file is freed
 instead of being fully opened.

The big one was that if the connect() call failed, it would loop
 forever (thanks to some code re-factoring, the loop was no longer
 advancing to the next address as originally designed).

On the first request made by a connection, the value it was testing
 might be uninitialized, so we can't guarantee it'll pass.

* Now supports HTTP/1.1 persistent connections with pipelining.
  This speeds up chain enumeration on large files by almost a
   factor of 2 over http, and by roughly a factor of 4 over https.
  The difference between http and https is now much smaller.
* Add timeouts to all the socket I/O.
* Estimate the number of available bytes to read and use it when
   making connection re-use decisions.
* Add support for https with proxies using HTTP/1.1 CONNECT
   tunnels.
* Fix TLS session re-use (it requires clean shutdown).
* Various other code re-organization and minor improvements.

Doing this in the source keeps them disabled when someone doesn't
 use one of our build systems.

It's helpful if we don't always read from the start of the array.
Thanks to Matti Virolainen for the report.

Making it exectutable doesn't put it in the path, so we have to
call it ./git-version.sh from inside the doc directory.

Jenkins didn't catch this because the $(shell ...) return code
doesn't seem to be propagated to make itself. It prints a warning
but doesn't abort. I don't know if there's a way to fix that.

Lets us call it directly without /bin/sh which saves space.

We use scripts like these with jenkins to do automatic build
and test runs on every commit. Keeping the build instructions
in the source repository itself makes it easier to track changes,
and easier to make atomic updates when something is moved in
the build system which breaks the jenkins runs.

This matches what the opus package does, and is consistent with
what the pkg-config files were already doing.

This is redundant now that we have PROJECT_NUMBER
substitution implemented. The file version is for
displaying per-file versioning from CVS, etc.

Saves an indent stop. Thanks to Tim for the suggestion.

This almost, but doesn't quite, copy the layout the opus
configure script uses.

* Tags were being freed even if not initialized if
   op_fetch_headers() failed on the first link.
  The logic for handling completely empty links would leak, also.
* Ignore fragment identifiers in http URLs instead of rejecting
   them.
* Get the current link in seeking_example's verify_seek when
   op_read_native() fails to return it for us.

Most of these were dead code left around from rewriting things and
 things llvm doesn't have enough global information to prove by
 itself.
The one real error was the missing NULL check in
 opus_tags_parse_impl().

Also use Libs.private for libm.

This displays the current raw position, PCM position, and
 instantaneous bitrate once per second of output.
This is useful when testing against long-running streams.

This makes it more useful for testing invalid files, where the goal
 is merely not to crash, assert, or trip valgrind warnings.

* Add Opus logo.
* Use @VERSION@ consistently.
* Fix some warnings/formatting exposed by a newer Doxygen versions.

The warning was correct, as far as it goes, but the use was
 actually harmless.

The previous comparison would have succeeded on things like
 "localhost123" instead of just "localhost".

* Guarantee pcm_start and pcm_end stay in range (not just move in
   the right direction).
* When we fail to find a page in the interval, back up by increasing
   chunk sizes just like op_get_prev_page_serial().
* Eliminate the special case for the last page in the interval.
* Force a straight bisection if we're backing up, but not decreasing
   the interval size rapidly enough, to limit the worst-case.
  This is guaranteed not to affect the first two iterations, so it
   has minimal impact on seeking in the normal case.

The autoconf build replaced the static Doxyfile with a Doxyfile.in
template. In the autoconf build, the config.status script applies
substitutions to generate the actual Doxyfile. This made the generic
doc/Makefile non-functional.

Apply similar substitutions directly with the makefile so we can
still build documentation outside the context of the autoconf build.

Also add a 'distclean' target to remove the generated Doxyfile.