Improve rotation stability

[nbraud]

• More accurate implementation of cos and sin, solves precision issue from Improve the Readme #87.
• Tighten tolerances on trigonometry stability tests.
• Expect exact results in unit tests, when applicable.

[nbraud]

PS: There's probably a performance hit in Vector2.rotate. Do we care?

[AstraLuma]

Can you categorize the performance hit?

[AstraLuma]

Most of these are non-functional stylistic changes.

[nbraud]

@astronouth7303 I think the commit whose style you find problematic is 5def680 ; it moved imports to ensure trig functions aren't used directly outside of Vector2.{trig,angle}.

[nbraud]

PS: Tell me when the review is OK, and I will clean up the history.

[AstraLuma]

That is the big one, although .trig() should be ._trig() to mark internal API.

[nbraud]

Yeah, I did that change too.

[nbraud]

OK, this is rebased & cleaned up

[nbraud]

Re: performance hit, I ran a pretty basic test (python3 -m perf timeit -s 'from ppb_vector import Vector2; v = Vector2(1, 0)' 'v.rotate(30)'):
On master: 1.58 us +- 0.08 us
On this branch: 2.42 us +- 0.14 us

I don't think we should care about a <1μs slowdown, even though that's 52% slower, considering that it gives less surprising results.

PS: Updated to use the perf package, which provides more accurate measurements and std deviation.
I also tried to test this under pypy, but it turns out it can't cope with type annotations, at least with the version packaged in Debian.

[nbraud]

What's the blocker on this?

[nbraud]

Unrelated CI failure, due to a missing rel_to in test_scalar_linear