- Fix error with usleep() being undefined.
 - Add support for 64-bit floating point formats for ALSA.

Fix an issue with the DirectSound backend where there'd be some leftover samples that are not sent to the application when a capture device is stopped.

  - Use binary semaphores as the synchronization primitive for the worker thread
  - Generalize the worker thread so as to avoid having different worker thread
    for each backend.
  - Make state changes atomic

General improvements bug fixing:
  - Fix an error where the DirectSound backend would immediately send the
    application a bunch of invalid audio data.
  - Have the ALSA backend request more audio data only immediately before it's
    ready to write it to the device.
  - General improvements on documentation.
  - Update readme.