Convert CubicBez::nearest to the poly-cool quintic solver

[jneem]

Currently CubicBez::nearest uses subdivision to QuadBez and then a cubic solver. This converts it to the quintic solver in poly-cool for a nice speedup (3000x for high accuracy!) and better robustness.

Fixes #446

Maybe it would make sense to propose moving poly-cool to linebender first? I wasn't sure...

cubic nearest point/0.001
                        time:   [126.51 ns 126.90 ns 127.31 ns]
                        change: [-78.774% -78.673% -78.577%] (p = 0.00 < 0.05)
                        Performance has improved.

cubic nearest point/0.000001
                        time:   [143.92 ns 144.16 ns 144.46 ns]
                        change: [-97.476% -97.469% -97.463%] (p = 0.00 < 0.05)
                        Performance has improved.

cubic nearest point/0.000000000001
                        time:   [162.17 ns 162.51 ns 162.88 ns]
                        change: [-99.971% -99.971% -99.971%] (p = 0.00 < 0.05)
                        Performance has improved.

This bumps MSRV from 1.82 to 1.85, which has more f64 methods in core and also supports edition 2024.

[Keavon]

Once poly-cool is used by Kurbo, I would recommend making a quick PR to finish up the tangent/normal to point family of functions, which would supersede #417 and add support for normals as well. Since that is low-hanging fruit to fill in some of Bezier-rs's extra functionality back in Kurbo where it seems noticeably missing.

[jneem]

Ok, I've updated this to use the newly-in-repo polycool.

[nicoburns]

I have no real insight into the math here. But this has been discussed several times and the consensus always seems to have been in favour of merging.