This is more flexible than the current approach:
- Gain can be changed if the aout has a custom (native) volume control.
- The callback approach does not break the encapsulation principles
  (so audio_output_t/module_need() could be used somewhere else).