We determine if a output frame is dropped by the arrival time and its
timestamp. However, the order of input buffers is different from
the display order of output buffers. So the expected arrival time is
wrong. Also, the base timestamp is updated after the first output
buffer dequeued. Therefore the first frame was always marked dropped.

In this CL, we calculate the expected arrival time by the count of
received frames, and set the base timestamp before checking the
arrival time of the first frame.

Bug: 155944233
Test: Run tast.arc.VideoDecodeAccelPerf.h264_2160p_30fps and check
      the first frame is not marked dropped

Change-Id: I1b34688ffa573016dd94d5fb3cc851462a25003d
(cherry picked from commit 4b5c13aa7e6a41d62c9f37338b493cd339548b19)