A2DP media task: Flush full TX queue on congestion

Currently, our TX queue does not allow for quick recovery of minor
congestion. As we only clear as much room in the queue as is needed to
fix the news frames needed for a given timer tick, any minor congestion
on the TX path can result in single frames to be dropped over an
excessively long amount of time.

From the "just rip the band-aid off" school of thought, drop the whole
queue instead to cause a single drop-out on congestion and catch up to
the live stream immediately, hoping that the stream recovers...

Bug: 29601962
Change-Id: I482c3cfef6d8b51143e54733eb2da5f70663636f
(cherry picked from commit d881c4b46776fe03a5861a3021bbe739b1f3cc58)