Revert "Sequester all Skia<->DL interactions into the skia sub-module (#40083)"

[jason-simmons]

This reverts commit b053d74.

[jason-simmons]

This was causing errors in the DanTup_tiler customer test (sorting/isometric golden image test)

See https://ci.chromium.org/ui/p/flutter/builders/try/Mac%20customer_testing/47476/overview

[flar]

LGTM

[flar]

The original PR was built from about 6 different commits so I can start with bisecting those to narrow down the cause...

[jason-simmons]

The sorting/isometric test diff reproduces with the first commit in the PR (288dfdb)

I can get the test to pass by reverting the changes to lib/ui/painting/canvas.cc and lib/ui/painting/paint.cc within that commit. But I haven't figured out which change within those files is causing the difference.

I did notice one change in behavior introduced in that commit. Before PR #40083 Paint::sync_to was checking flags.is_stroked(builder->getStyle()) to decide whether to set stroke properties. It was doing this even if flags.applies_style() was not set. So the state of the builder prior to calling Paint::sync_to could affect what sync_to does.

PR #40083 changes this so that the builder's current state is not available to Paint::paint. Paint::paint sets stroke properties based only on the paint data passed from Dart. This seems correct.

But in any case, it does not look like this issue is affecting the sorting/isometric test.

[jason-simmons]

I found the issue - in Canvas::drawImageRect, the call to builder()->DrawImageRect should pass false as the last argument.

(It's currently passing the enum SkCanvas::kFast_SrcRectConstraint, which is intended for the old drawImageRect method and is being implicitly cast to enforce_src_edges = true)

[flar]

And a quick grep shows a lot more occurences of that flag in the DL code. I'll follow up and see if any of the others are wrong. It's annoying that the compiler allows implicit conversion between a bool and an enum. Grrr...