msm: mdss: fix asynchronous rotation of layer bigger than line buffer

If MDSS HW has more than 2 rotators and if the layer width is greater
than line buffer width then layer is divided into two and each half
can be rotated independently and asynchronously. Current implementation
sets up the configuration of both halves independently but the rotation
of second half is triggered only after first half is finished. Fix this
issue by making writeback_kickoff unblocking call and wait for the
rotation completion after all the rotations in the current session are
queued.

Change-Id: I90ff7feab7979037e596a3b8d8c9e6c138574e94
Signed-off-by: Ujwal Patel <ujwalp@codeaurora.org>