Previously, the streams' offsets needed to be pre-adjusted with the
interleave; now assumes all streams start in the same offset (first
stream).
This simplifies short last interleaves (SCD/P3D) and makes
layout_mpeg_custom unnecessary (also allows theoretical variable-sized
interleaves).
To support future MPEGs of uncertain layout and frame variations (namely
EALayer3) the code has been restructured: mpeg_decoder does stream
procesing and decoding, while mpeg_custom_utils_x does init and parsing
(write to data buffer + update offsets), per MPEG subtype. Internals
have changed but still gives byte-exact results.
AHX has been adapted to this format as a test. Some modes
(P3D/EALayer3/LyN/AWC) are defined for later development but will fail
on init ATM.
Rarely used, frame sample rate/channels are now exposed, and FFmpeg will
end up playing any MP3 anyway.
Encoders may modify sample rate to adjust looping to frame boundaries,
too.
Also removes VBR check from metas; it's only needed because
mpeg_bytes_to_samples doesn't work with VBR, but now it'll return 0,
which would be validated later.
