Experiment with creating a vello-cpu-backed renderer

[LaurenzV]

Not meant for review, this is just experimenting to see how far we can get and what issues arise.

So far, some of the main missing features are:

• Non-isolated blending
• Non-premultiplied gradient interpolation

[RazrFalcon]

Is vello-cpu a tiny-skia replacement?

[LaurenzV]

I mean, depends on your definition of "replacement", but it's also a CPU-only 2D renderer. :)

[LaurenzV]

It's quite a bit faster than tiny-skia though, see also: https://laurenzv.github.io/vello_chart/

x86 benchmarks are still missing, though.

[RazrFalcon]

Well, ARM benchmarks are unfortunately useless, since I have never had time to optimize ARM. x86 should be more interesting, but not mind blowing.

Also, tiny-skia being slower than cairo and even raqote is very strange. Maybe something have changed recently, but tiny-skia was consistently faster than cairo. Well, at least 5 years ago when I wrote it.

[LaurenzV]

Yeah, I was wondering the same tbh. I will definitely do more validation once I get to it (the benchmarks are for my master's thesis, so I definitely wanna make sure I got everything right), but I wouldn't know what I did wrong.