Fix vorbis playback

This fixes a bug where the wrong position is being retrieved for the
capture device.

This change makes it so reading and writing is done on the smaller of
the period sizes between capture and playback devices.

The new rule is to start playback when the entire buffer is filled
rather than when the first period is filled in duplex mode. Normal
playback mode will still start the device when the first period is
filled.

This commit attempts to address a crackling issue in full-duplex mode.

* When the capture buffer becomes over-filled it will discard some
  frames to prevent it from straddling the capacity of the internal
  buffer and causing excessive overruns.
* A fix has been implemented to the function that retrieves the number
  of available frames in the device's internal buffer.
* Low-latency shared mode (IAudioClient3) forces small buffer sizes so
  a change has been implemented which bypasses low-latency shared mode
  when a large buffer is specified.

The APIs that have changed are those that take an output pointer and an
input count in the format of "some_read_function(handle, pBuf, count)".

The updated APIs include:
  * mal_decoder_read_pcm_frames()
  * mal_pcm_converter_read()

Some APIs are scheduled to be replaced in a future version so I have
decided not to replace this at this time since it will be a waste of
time to do so.

This is the same as the normal ring buffer except everything is done in
terms of PCM frames instead of bytes. This is useful to simplify the
case of using the ring buffer for storing PCM audio data.