msm: vidc: handle dynamic buffers in pendingq during flush

In dynamic buffer mode, if flush was received before
stream-on, driver would have incremented the ref count
for output buffers, but not yet sent to firmware. So,
after flush done, there would be no fbd from firmware
and the reference to these buffers would not be removed.
So when the buffers are dequeued, driver fails to
unmap and free the output buffers.
Subsequently, after flush done, when client queues
the same output buffers again to driver, the ref
count is increased and becomes 2. This results in
the driver holding the buffers with itself, thereby
entering in to a deadlock situation and starving the
firmware of output buffers. This results in a video
freeze.

b/17647776

Change-Id: I0c3d13124bd48b900e56e14666eaed2f0e7ee9d6
Signed-off-by: Arun Menon <avmenon@codeaurora.org>