Add extremely basic RDO

On a rescaled 192x64 version of akiyo_qcif this reduces file size by
~10% for a ~.3 dB loss on PSNR.

- Only chooses between DC_PRED and H_PRED at the superblock level
- Distortion is just SSE
- Uses floating-point math for no reason

Test the entropy encoder

intra_mode_ind is no longer used

Leverage the aom reader.

Make the --output parameter mandatory.

Fix the example command line parameter.

Drop the --release parameter from cargo so it builds faster.

Move pred_dc_4x4 to testing namespace

Since it is only available for testing.

Add Travis build status badge.

Fix command line

Change the rav1e command line to take an option -o OUTPUT to match every
 other encoder.
If no output is specified, it will write to stdout.

The encoder now prints status to stderr instead of stdout.

Use a Box<Write> instead of File for the output_file field of
 EncoderFiles.
This matches the input_file: Box<Read> field.

Implement a pure-rust pred_dc

And add some tests.

Fix #2: Enable Travis

Add (but do not use) H_PRED support

If chrome uses H_PRED then the transform type is DCT_ADST, but we were
always using DCT_DCT, we now lookup the correct transform type using
`exported_intra_mode_to_tx_type_context`.

Fixing this also requires using the correct scan table for each
transform type. This is handled using the lookup table
`av1_intra_scan_orders`.

The encoder still uses DC_PRED everywhere until we get RDO in.
This required turning off ext-intra to work.

Note that this only works for luma blocks, H_PRED support for chroma
blocks is fixed in a latter commit in this PR.

- Mode::pred -> Mode::mode since we use this in function like
write_intra_mode, etc.
- SubMode::pred -> SubMode::mode

- Mode::new() -> Mode::default() since "new" is a bit vague.
- SubMode::new() -> SubMode::default()

Use correct AV1 fourcc