Allocation callbacks must now implement onRealloc() themselves. This is
no longer emulated by miniaudio.

With this change, ma_decoder should now be fully consistent with other
data sources.

This makes ma_decoder_read_pcm_frames() consistent with other data
sources. It now returns a result code and outputs the number of frames
read via an output parameter.

This is now consistent with all other data sources:

  * Returns a result code
  * Length is returned via an output parameter

Use the `encodingFormat` variable in the config instead.

This also renames the `resourceFormat` variables used by the encoder to
`encodingFormat`.

This was used with the old logging level system which is no longer
being used.

This has been replaced with MA_LOG_LEVEL_DEBUG.

This was specific to devices, but now that miniaudio has expanded it's
no longer useful. The new logging system is more generic and useable by
parts outside of devices and contexts.

This has proven to be exceptionally un-useful in practice and the
maintenance cost just isn't worth it.

This might come back later, but for now this is no longer a thing.

This now takes an input and output buffer rather than performing the
clipping in-place.

The previous version only allowed in-place clipping, whereas the new
one supports clipping into a separate buffer. The input and output
buffers can point to the same buffer in which case the clip will be
performed in-place.

This has been replaced with ma_silence_pcm_frames().

This removes the `pBufferOut` parameter from the following APIs:

  * ma_rb_commit_read
  * ma_rb_commit_write
  * ma_pcm_rb_commit_read
  * ma_pcm_rb_commit_write

These are replaced with ma_standard_sample_rate_min/max.

This fixes an issue where excess channels are not heard when the sound
source has more channels than the listener.

This commit also fixes a bug where panning is incorrectly getting
applied to non-directional channels, such as mono and LFE channels.