Without escaping, some markdown renderers convert double dashes to
em dashes

Co-authored-by: Wolfgang Stöggl <c72578@yahoo.de>

Fixes https://github.com/xiph/flac/issues/792

Credit: Oss-Fuzz
Issue: N/A

Also, make msys2 action find git executable

As https://github.com/xiph/flac/issues/784 notified me of
undefined behaviour going undetected by the fuzzers, I ran the test
suite with the full undefined behaviour sanitizer enabled.

There is one bit of undefined behaviour that I cannot fix,
FLAC__metadata_get_picture allows a picture type of -1 to mean all,
but this isn't part of the enum.

Fixes https://github.com/xiph/flac/issues/784



Co-authored-by: braheezy <11700503+braheezy@users.noreply.github.com>

Credit: Oss-Fuzz
Issue: https://issues.oss-fuzz.com/issues/42537590

This slightly speeds up the encoder by reducing the number of
syscalls, but can result in a 30% speedup in extreme cases, mostly
when using extremely small blocksizes.

Co-authored-by: Martijn van Beurden <mvanb1@gmail.com>

Some subdirectories had a higher minimum version specified. This
is now 'propagated'. Also, compatibily with newer CMake versions
is now explicity stated.

This should reduce the number of timeouts. In case the decoder
keeps rewinding, check 4 of the 8 CRC8 bits. If they match, keep
rewinding.

With this approach, the fuzzer doesn't get 'faster' then plain
libFLAC, but fuzzing isn't affected too much by the CRC8 check.

Time will tell whether the fuzzer finds an input triggering
repeating rewinds, expensive processing ánd those 4 CRC8 bits
check out. If it finds such an input, more CRC8 bits need to be
checked.

Credit: Oss-Fuzz
Issue: N/A (local run)

See https://github.com/xiph/flac/pull/765 for report.

Growing up to the max allowed size took roughly 4000 reallocations,
which can be expensive. Now grow exponentially, which requires
about 30 reallocations up to max size.

Fixes https://github.com/xiph/flac/issues/767

The max number of errors and number of output bytes are limited

As files with a changing number of channels or bit depth are rare
and difficult the handle anyway, improve fuzzer performance with
them by limiting realloction of the decoding buffer

Credit: Oss-Fuzz
Issue: N/A (local run)

... and use realloc of linkdetails more efficiently. This should
reduce the number of fuzzer timeouts for fuzzer_seek

Credit: Oss-Fuzz
Issue: https://issues.oss-fuzz.com/issues/379254072

See https://github.com/xiph/flac/issues/761