lacros: create one fence for raster images.

When delegated compositing is enabled, rasterization
is done to SharedImageBackingGlImage, which creates
a glFence at the end of each task. That results in
creation of big number of fences, which takes time and might
have some overhead depending on a driver (in this particular
case, we are fixing the perf problem with i965 driver) - it
can easily be >50 fences.

Thus, to avoid that, we can create just a single fence
when all raster tasks are executed. The ideal place to do
that is during OutputPresenterGL::ScheduleOverlays call and
pass that fence to the GLSurface as a fence for the current
frame, which must be used for those overlays that do not have
a fence. Then, when WaylandFrameManager receives a CommitOverlays
call, it checks if the fence has already been signaled, and
destroys that. If it hasn't been signaled, a fence is cloned
and set only to those subsurfaces, whose overlay config doesn't
have a separate fence.

By doing this optimization, we reduce the time spent on
creation of fences. MotionMark.Images benchmark (with
only 1 cpu available and limited to 2ghz) produces
~38 FPS instead of ~31-32 FPS. Please note this is only valid
for the case when delegated compositing actually fails and
compositing is done on the lacros side (this happens because
number of quads is limited to 64).

However, there are some other performance problems with the
DelegatedCompositing itself, which results in about ~20FPS using
the same test case. This will be tackled in another CL.

Cq-Include-Trybots: luci.chromium.try:lacros-amd64-generic-rel,lacros-arm-generic-rel,gpu-fyi-try-lacros-intel-rel
Bug: 1254033, 1296214
Change-Id: I76c797eda63dee546c395b4c2cf8b66406ca8649
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/3483799
Reviewed-by: Vasiliy Telezhnikov <vasilyt@chromium.org>
Commit-Queue: Maksim Sisov <msisov@igalia.com>
Cr-Commit-Position: refs/heads/main@{#984337}

• cc/raster/gpu_raster_buffer_provider.cc[diff]
• components/viz/service/display_embedder/output_presenter_gl.cc[diff]
• gpu/command_buffer/common/shared_image_usage.h[diff]
• gpu/command_buffer/service/shared_image_backing_gl_image.cc[diff]
• gpu/command_buffer/service/shared_image_factory.cc[diff]