2.0 perf updates

[davepagurek]

Resolves #7539

• Switch to shallow cloning in push/pop + copying before modification of state values
• Refactors push/pop to only update changed properties
• Creates a States class to keep track of what values have been set
• Switches from states.key = value to states.setValue('key', value) because using a getter/setter was noticeably slower
• Makes sure p5.Matrix values are always Float32Arrays
• Removes structuredClone from color maxes
• Puts cached stroke/fill style values into states to avoid having to read them from the canvas again on pop

To do:

• Make serialize for fills faster

[davepagurek]

@limzykenneth This should be testable now!

I had to update some screenshots for one text assignment test, but they were only failing on CI. Looks like some text was very close to wrapping and some change made it just different enough to not wrap on CI. I updated the font size for that test to not be right on the edge like that, which feels less clean than I'd like, but... I'm no closer to figuring out what shall detail changed than I was an hour ago :')

[davepagurek]

ok TIL that test is failing on dev-2.0 too currently. I think that test flake is unrelated to this PR, and possibly even caused by unrelated CI dependency changes? but anyway that makes me feel more confident that this PR didn't break it!

[dhowe]

still seeing this issue on PRs

[limzykenneth]

I was thinking since the class have some similarity to Map would it be worth to implement it as an inherited class from Map but I think that probably will complicate things and not give much benefit.

It seems the test is also passing here now so I guess if we merge this it should fix the test on the branch and other PRs as well?

[davepagurek]

@limzykenneth feel free to merge when the color serialization is ready!

[limzykenneth]

Still trying to figure out a more comprehensive cache for color object in general. That's because with each color object created, it will need to be serialized once first before it will be memoized but with each call to any function that takes simple number arguments, it will create a one time used color object which erases the memoization benefit.

If you have an idea around here do share, I'll probably spend a bit more time on it and if I still have nothing I'll merge this for now.

[limzykenneth]

For some reason I'm not getting the full 60 fps on my linux PC (getting around 20 fps) not even with 1.11.3 (which gets about 30 fps), while on my Macbook Pro I'm able to get full 60 fps. Surely are Macs that good now?

Anyway, that's beside the point. We can continue looking into other avenue for more optimizations but if it isn't too hard to get 60 fps on the original sketch it should be ok for now.