The old formulation
* wouldn't work if there were ID3 tags at the end
* would choose whether to background-refill the inbuf
based on a check before skipping to the next sync word, which
could be incorrect.
I think it was aspect "B" that ended up triggering the erroneous EOF
problem fixed in the prior commit. This would depend on specific data
sizes and offsets occuring in the file such that a read would be
scheduled but then the buffer would be filled and left 100% full by
find_sync_word(). It's just lucky(?) that a particular person produced
such a file, and/or many files produced by Audacity have those
characteristics.
df5f7bc765