sf-latency: Explicitly time SKIA_GL_THREADED

Bug: 193240340
Test: This

My original intent for this benchmark was to time drawLayers, without
worrying about threading overhead. But using SKIA_GL means that we do
not execute mapExternalTextureBuffer ahead of time, so we have to do
this work each frame. mapExternalTextureBuffer should be tracked by a
separate benchmark.

Change-Id: I0d58fdf67a6322f0bbc04835ab1cdb08b73bcf7b