This is mainly for performance. Without bounds, skia may execute some
code to find a finite bounds, and the bounds may be too big for the
SaveLayerOp which may lead other cost.

Cluster telemetry showed that this CL can improve raster_time by 8%.
https://ct.skia.org/results/cluster-telemetry/tasks/chromium_perf_runs/wangxianzhu-20180318183406/html/index.html

Pin-point showed 20% improvement for raster_time in
rasterize_and_record_micro.top_25.
https://pinpoint-dot-chromeperf.appspot.com/results2/14f80e29440000

Another purpose is for some skia bug causing crbug.com/816854. In the
bug, a small chunk with opacity without bounds in SaveLayerOp may
cause ghost rasterization far from the target area.

Bug: 803867,816854
Cq-Include-Trybots: master.tryserver.blink:linux_trusty_blink_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_layout_tests_slimming_paint_v2
Change-Id: Ife282e44b09fb2978e850e4c8cd6ca4e0150c857
Reviewed-on: https://chromium-review.googlesource.com/966699
Commit-Queue: Xianzhu Wang <wangxianzhu@chromium.org>
Reviewed-by: vmpstr <vmpstr@chromium.org>
Cr-Commit-Position: refs/heads/master@{#544561}