Reland ImageFiltered bounds fix

[flar]

The original fix in #16960 was passing the filtered bounds to the child savelayer which means 2 things:

• when the filter grew the bounds, the child was painting into an unnecessarily large temporary surface before the filter grew the result again - which means a lot of transparent pixels would be in the source of the filter and the filter would have more useless work to do.

• when the filter shrunk the bounds, the child was given a temporary surface that was too small to draw itself into and so the output was cropped.

Fixes flutter/flutter#51978

[liyuqian]

It would be nice to add a unit test that would fail if 2f971c8 isn't patched.

[flar]

Forgot to update the unit tests in the engine. On it...

[flar]

And, for the record, I also reran the framework golden tests for this widget with this local engine build and they passed.

[flar]

It would be nice to add a unit test that would fail if 2f971c8 isn't patched.

The existing unit tests that were landed in the original fix started failing when this was patched. I had to update them to reflect the new relationship between filtered and child bounds. So they are essentially asserting that the child and parent bounds meet the constraints now implemented.

[flar]

For added test coverage I updated the new SimpleFilterBounds test with the code we use to verify the mock drawing calls.

Note that this created what looks like a strange diff because the tail end of the brand new SimpleFilterBounds test is character by character identical to the tail end of the old existing SimpleFilter test and so it confused which function owned the last 10-12 lines.

[flar]

Ping

[liyuqian]

LGTM with a nit.