Stop using pendingIO in read/writeSurfacePixels (take 2)

This changes the implementation to always flush for these operations.
Once partial DAG flushes are implemented this behavior will be better than the old method.

The following two CLs have suppressed all the failures that caused the first try to be reverted:

https://skia-review.googlesource.com/c/skia/+/206069 (Blacklist two more GMs for gltestthreading)
https://skia-review.googlesource.com/c/skia/+/206171 (Disable the TextureStripAtlasManagerColorFilterTest on MoltenVK)

TBR=bsalomon@google.com
Change-Id: Ie2d09ebcba03f9987a42038605801f7341fc72a6
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/206177
Reviewed-by: Robert Phillips <robertphillips@google.com>
Commit-Queue: Robert Phillips <robertphillips@google.com>
1 file changed