TODO:
- directsound: incomplete and somewhat broken
- osx: understand why AudioQueueGetCurrentTime can return negative mSampleTime
- test (and fix) sub-prefill size data playback
- report stream delay instead of position; leave position calculation to user
- capture support
- capture and output enumeration and configuration
  - also expose default hardware config to allow decisions on speaker layout
- prefill occurs at different times in each backend:
  - pulse prefills async off worker thread after init
  - coreaudio prefills during init
  - alsa prefills async after start
- expose configured prefill size; may differ from requested latency
  - solved by exposing stream delay
- xruns may occur in user callback but also in audio hardware
  may need to expose details of hardware xruns to user api
- document thread safety
- document which calls may block, and when effects take effect
- document what's permissible inside callbacks
- implement basic channel mapping for surround
  - vorbis has documented mapping based on channel count (if mapping type ==
    0) -- http://xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-800004.3.9
    1 -> M
    2 -> L, R
    3 -> L, C, R
    4 -> L, R, RL, RR
    5 -> L, C, R, RL, RR
    6 -> L, C, R, RL, RR, LFE
    7 -> L, C, R, SL, SR, RC, LFE
    8 -> L, C, R, SL, SR, RL, RR, LFE
   >8 -> application defined
  - wave files with channel count only
    3 -> L, R, C
    4 -> L, R, RL, RR
    5 -> L, R, C, RL, RR
    6 -> L, R, C, LFE, RL, RR
    7 -> L, R, C, LFE, RC, SL, SR
    8 -> L, R, C, LFE, RL, RR, SL, SR
  - wave files with WAVE_FORMAT_EXTENSIBLE have explicitly mappings, can
    extract these
- implement configurable channel mapping