video: msm: Fix corner cases of "read twice" error

Fix several cases where histogram's completion structure is completed
multiple times which causes the next histogram read to force the histogram
software state machine into believing that the histogram is always being
read by someone.

When the timeout in mdp_do_histogram times out and then soon after the work
that was being waited on completes, the next do_histogram call that is
processed will not wait (and thus will not have its hist field set to
NULL). This will cause all subsequent do_histograms to think that the
histogram is currently being read, and will cause the "read twice" message
to flood the kernel logs.

This error is fixed by always initializing the completion structure before
waiting for the histogram to complete. That way if the completion happens
after the timeout, instead of causing the error, the completion's internal
counter is reset to zero, which prevents the next histogram read from
completing without waiting.

Change-Id: Ic1bc4b8688cd191d226de1df15d389de09996106
Signed-off-by: Padmanabhan Komanduru <pkomandu@codeaurora.org>
Signed-off-by: Carl Vanderlip <carlv@codeaurora.org>