* finalize the API:
  - which PCM output format (s24,s32,float...) ?
  - any special needs for the extensions ?

* implement missing features:
  - check output gain! I couldn't find anything in the spec about this so
    right now I'm assuming the output of the 32 subbands QMF is supposed to
    be in 16 bits/sample form but this seems very dubious.
  - fix downmix (user proper channes coefs, move downmix before QMF)
  - sum/difference audio channels
  - joint channels coding
  - dynamic range
  - dialog normalization
  - check audio output accuracy (right now the main issue is the output gain)
  - extensions (96k, extra channels)

* optimize:
  - don't use doubles everywhere
  - integerize everything ? (main difficulty should be the QMF and LFE FIR)
  - faster 32 subbands QMF

* reliability:
  - make the parser more tolerant to faulty bitstreams
  - more sanity checks
  - crc checks
  - regressions tests

* code cleanup