It was decoding one 28-sample subframes and blocked layout handled
frames, doing some voodoo to skip CD sectors.
Now decodes one 28*8-sample frame and blocked layout handles CD sectors.
Decoding is byte-exact vs before.
changed _noninterleaved functions to _int and changed related descriptions appropriately
added an _int version of SDX2 and reverted to pre-322 behavior for normal deocder
made the various get_high_nibble globals all be the same vgmstream->get_high_nibble for thread safety and code consolidation (this required some changes to make decode functions take VGMSTREAM rather than VGMSTREAMCHANNEL)
git-svn-id: https://vgmstream.svn.sourceforge.net/svnroot/vgmstream@323 51a99a44-fe44-0410-b1ba-c3e57ba2b86b