When this is used, the data callback should at some point call
ma_engine_read_pcm_frames() in order to do some processing.

This is optional and is fired at the end of each call to
ma_engine_read_process_pcm_frames(). The callback will be passed the
processed audio data so they can do their own processing such as
outputting to a file or whatnot.

The callback is configured via the engine config.

With this change, dr_wav is now namespaced with "ma" which means dr_wav
can now be used alongside miniaudio.

In addition, some duplicate code has been removed, such as sized types,
result codes, allocation callbacks, etc. which reduces the size of the
file slightly.

This should address the following public issue:
  https://github.com/mackron/miniaudio/issues/673

miniaudio is updating it's amalgamation of dr_wav, etc. so that it's
all namespaced with "ma" which will make the amalgamated versions of
dr_libs entirely independent. There's no longer any need to mention
the decoding backends.

Documentation regarding stb_vorbis is removed so as to discourage
new users from using it. Support will not be removed until a
replacement Vorbis decoder can be amalgamated, but new users should
instead be guided to the libvorbis custom decoder in the extras folder.

These compilers do not support noinline.

  * Sized types will now use miniaudio's types.
  * Architecture macros now use miniaudio's macros.
  * The c89atomic namespace has been renamed to ma_atomic which makes
    it so c89atomic can be used alongside miniaudio without naming
    conflicts.

Public issue https://github.com/mackron/miniaudio/issues/673

This is in preparation for some amalgamation improvements.

This commit makes it so SetFilePointer/Ex() are dynamically loaded at
runtime which allows miniaudio to branch dynamically based on available
support.

This is necessary because versions of Windows prior to XP do not
support the Ex version which results in an error when trying to run the
program.

Public issue https://github.com/mackron/miniaudio/issues/672

Since the calling thread is waiting anyway, it's better to just do the
processing on the calling thread rather than posting it to the job
queue and waiting. This ensures the calling thread stays busy which
will improve performance, but it also makes it so the calling thread
doesn't get stalled while already-queued jobs are getting processed.