Each client should add all primary surface ids into activation dependencies on context loss.

I actually don't believe resetting fallback SurfaceId in the browser process do much other than avoiding missing surface errors. If the primary is not in activation dependencies it won't trigger surface sync.

Yea, we can also just skip the error in SurfaceAggregator...

I'm just going to get rid of the error case in SurfaceAggregator.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/aec1f1576ed70cbdb40837814b6cf9c09e1f5bd0

commit aec1f1576ed70cbdb40837814b6cf9c09e1f5bd0
Author: Fady Samuel <fsamuel@chromium.org>
Date: Wed Jul 25 21:05:11 2018

SurfaceAggregator: Remove Missing Surfaces report

With OOP-D, after a Viz process crash, clients will start
resubmitting CompositorFrames in no paritcular order and
some references may not yet exist at aggregation time. In that
transient period, we shouldn't spew errors to console or to UMA
because this is an expected transient period. This CL simply
removes that error state. In the absence of a fallback surface,
we use the default background color of the SurfaceDrawQuad to fill
in the surface's area.

Bug:  865597 
Cq-Include-Trybots: luci.chromium.try:android_optional_gpu_tests_rel
Change-Id: I45d10c9a1880ea9df04dd7a95ef76048abc4a830
Reviewed-on: https://chromium-review.googlesource.com/1150574
Reviewed-by: Alexei Svitkine <asvitkine@chromium.org>
Reviewed-by: Saman Sami <samans@chromium.org>
Commit-Queue: Fady Samuel <fsamuel@chromium.org>
Cr-Commit-Position: refs/heads/master@{#578060}
[modify] https://crrev.com/aec1f1576ed70cbdb40837814b6cf9c09e1f5bd0/components/viz/service/display/surface_aggregator.cc
[modify] https://crrev.com/aec1f1576ed70cbdb40837814b6cf9c09e1f5bd0/components/viz/service/display/surface_aggregator.h
[modify] https://crrev.com/aec1f1576ed70cbdb40837814b6cf9c09e1f5bd0/tools/metrics/histograms/histograms.xml

FYI this is still broken. It's no longer reporting missing surfaces but the OOPIFs don't necessarily show up until something else (eg. mouse move) triggers a new DrawAndSwap().

My guess is this is because the renderer surface can't add a reference to a OOPIF surface that doesn't exist. The renderer doesn't need to submit another CF because nothing changes when it gets the OnFirstSurfaceActivation() message for the OOPIF surface, so nothing changes and nothing triggers the next DrawAndSwap().

Saman fixed this in another bug so I'm marking as FIXED.