render: Stroke paths with tiny-skia-path

[torokati44]

This is supposed to fix #7565 by not using lyon's StrokeTessellator - instead, for strokes, converting the Ruffle path to a tiny_skia_path::Path first, then running a tiny_skia_path::PathStroker on it, then converting that to a lyon path, and tessellating it as a fill with a non-zero fill rule.

It's dumb and probably slow, but hey, it seems to work.

Fixes #13656, and fixes #13658 too.

[Dinnerbone]

probably slow

Any metrics on this already or do you want me to benchmark something?

[torokati44]

No metrics yet at all. I am currently just surprised that this approach even works as well as it appears to do (and first try, thanks Rust!).

I plan to do some benchmarking soon, but if you find the time and initiative, more eyes on it can't hurt.
Although I don't think this is in its final form yet.

Related: I suppose a wireframe rendering mode could come in handy for debugging/visualization/coolness purposes... I know it's not as simple now as back in the old GL_LINE_LOOP days, but any rudimentary solution would be fine for this.

[torokati44]

Somewhat related, for debugging/verifying/inspecting purposes: #11880

[torokati44]

One small problem with this is that tiny-skia-path does not support different line cap styles at the start and the end of a path. I can try to add and PR this, but no promises. It's not that often needed, AFAIK only available through the Graphics drawing API (or something like that), and not from SWF tags.

Also, the exact dotting/dashing algorithm might be different (see the updated avm2/graphics_direct_commands expected test output) but I don't find this to be that big a deal.

And finally, this way of making a stroke mesh might, in some rare cases (such as in the linked z0r.de loop) produce highly suboptimal meshes, with lots of unnecessary internal vertices. While we could try to be clever and throw those away, I don't think the additional CPU overhead of the already slow tessellation justifies the handful of vertices saved for the GPU. I could be wrong though...