[CELT-dev] Crash bug in CELT plc

[Timothy B. Terriberry]

> Actually, I think the "correct" thing to do is for mode setup to give
> you 160-sample shorts instead of 80-sample shorts (and thus LM will
> never be 2).

Sorry, spoke before thinking. I was on crack. I think your two 
suggestions are actually fine. The loop before that which fills in X[] 
with the content of the PRNG should also probably stop at 
(st->mode->eBands[st->mode->effEBands]<<LM), but denormalise_bands() 
should clear the tail of the spectrum anyway, so that doesn't actually 
matter.