sf-latency: Wait on fence after each drawLayers call

Bug: 193240340
Test: this

Previously, the benchmark simply passed the fence to the next call,
meaning that the next drawLayers could start before the GPU was
finished. Explicitly wait for it inside the benchmark, so that we'll
time both CPU and GPU work.

Add comments describing such.

Change-Id: I7f34636d6b3701b71eaa8d8435e947e258d2fa67