textToPoints perf improvement

[davepagurek]

textToPoints is known to be sort of slow because getting relatively even sampling in Cartesian space along a Bezier curve isn't computationally easy to do. The main change here is that I've replaced the implementation we currently have with one I've been working on on the side for a while.

I was using this test case to measure performance:

const points = font.textToPoints('p5*js', 0, 0, { sampleFactor: 2 });
p.beginShape(p.POINTS);
for (const { x, y } of points) p.vertex(x, y);
p.endShape();

Sampling density	Current implementation	Updated implementation
2	15 FPS	60+ FPS
70	0.4 FPS	15 FPS

On p5 1.x, this would give me 15fps on my 2021 M1 Macbook Pro. After these changes, this test case runs at 60fps.

To test the speedup, I increased the sampling density until it went back down to 15fps. I found I had to crank the density up to 70 to hit the same speed as before, so this seems to be a 350% speedup on this test case.

Under the hood, the main difference is that rather than using iteration to narrow down on the point at a given length for every sample, we can, upfront, calculate a much smaller number of coarse samples at evenly spaced intervals, and then essentially use these as a skip list to quickly narrow down on the target point.

The other update here is that I've reverted the default sampleFactor value to what it is in 1.x (I didn't realize before that this was an existing API -- the small adjustment I made to the defaults earlier isn't important enough to warrant a breaking change.)