Change mLayersWithQueuedFrames to be an unordered_set

It was possible for mLayersWithQueuedFrames to grow unbounded for very
long GPU work, which causes additional calls into the kernel in
latchBuffer() to repeatedly call into the kernel to check the status of
the same fence. Instead, mLayersWithQueuedFrames should be a set so that
duplicate layers are skipped.

Fixes: 167300366
Test: systrace
Change-Id: I0c54ea6265ce901192e430c8f446a04547e1b833