There are 4 scenarios to consider when capturing audio from a channel
with an audiohook:

 1. There is no rx and no tx audio, so return nothing.
 2. There is rx but no tx audio, so return rx.
 3. There is tx but no rx audio, so return tx.
 4. There is rx and tx audio, so mix them and return.

The file passed as the primary argument to MixMonitor will be written to
in scenarios 2, 3, and 4. However, if you pass the r() and t() options
to MixMonitor, a frame will only be written to the r() file if there was
rx audio and a frame will only be written to the t() file if there was
tx audio.

If you subsequently take the r() and t() files and try to mix them, the
sides of the conversation will 'drift' and be non-representative of the
user experience.

This patch adds a new 'S' option to MixMonitor that injects a frame of
silence on either the r() side or the t() side of the channel so that
when later mixed, there is no such drift.

Change-Id: Ibf5ed73a811087727bd561a89a59f4447b4ee20e