Enable the atlas path renderer for GrAAType::kCoverage

This will definitely cause regressions on a whole slew of benchmarks
that were carelessly regurgitating cached bitmaps until the system
memory was blown. But the well defined memory limits of the atlas and
resilience to animation are worth it. If a client still wants this
type of caching they can always render their own bitmaps using Skia
and cache them.

Bug: chromium:928984
Bug: skia:12258
Change-Id: I277f50cb9192f1cf5fe8d4b7f629abe72432150a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/433917
Commit-Queue: Chris Dalton <csmartdalton@google.com>
Reviewed-by: Brian Salomon <bsalomon@google.com>
diff --git a/src/gpu/ops/GrAtlasPathRenderer.cpp b/src/gpu/ops/GrAtlasPathRenderer.cpp
index b9ac4a6..66b2f63 100644
--- a/src/gpu/ops/GrAtlasPathRenderer.cpp
+++ b/src/gpu/ops/GrAtlasPathRenderer.cpp
@@ -220,9 +220,13 @@
     SkASSERT(!args.fHasUserStencilSettings);  // See onGetStencilSupport().
 #endif
     bool canDrawPath = args.fShape->style().isSimpleFill() &&
+#ifdef SK_DISABLE_ATLAS_PATH_RENDERER_WITH_COVERAGE_AA
                        // The MSAA requirement is a temporary limitation in order to preserve
                        // functionality for refactoring. TODO: Allow kCoverage AA types.
                        args.fAAType == GrAAType::kMSAA &&
+#else
+                       args.fAAType != GrAAType::kNone &&
+#endif
                        !args.fShape->style().hasPathEffect() &&
                        !args.fViewMatrix->hasPerspective() &&
                        this->pathFitsInAtlas(args.fViewMatrix->mapRect(args.fShape->bounds()),