msm: mdss: push wait for fences and ping pong done as late as possible

All the configuration up until MDP flush is double buffered. This means
we are able to do all hardware programming without settings taking
effect until flush happens. Perform all programming, wait for fences
and wait for previous frame is done just before the flush in order to
reduce the time between last frame done and kickoff of next one.

Change-Id: I1fd0c7a65309a3b148c8efef61819bfe9a248f29
Signed-off-by: Adrian Salido-Moreno <adrianm@codeaurora.org>