Proposed benchmark infra, adds collision detection benchmark, experimental performance improvements

[cspotcode]

This is a draft meant to accompany Discord conversation, but is not meant to be merged. Individual commits will, I hope, be useful and can be cherry-picked into separate pull requests.

The reports are confusing, I need to clean them up. But basically, every commit on this branch was checked out, and then the benchmark was run. The results show most recent commits first, older commits at the bottom. If you see a noticeable change in performance between one commit and the previous commit, then you know it made things faster/slower, and you can look at the changes within that specific commit. We can choose the commits that were most beneficial and extract them into a separate pull request.

If you try to run these benchmarks yourself, you'll need to tweak bench.sh and install hyperfine. When I come back to this, I'll add proper instructions.

Over two benchmark runs, these are my results.

The big improvements came from:

• d5cbb312c4c8a83388acfb96f83a683a527d87fc
• 055c42413039aa40c458e93663c60053f1dd66e9

Command	Mean [s]	Min [s]	Max [s]	Relative
python -m benchmarks.collisions.bench d93ba54ace7feb49cd949856a28d4e88977b1265	17.273 ± 0.263	17.087	17.458	1.12 ± 0.02
python -m benchmarks.collisions.bench 055c42413039aa40c458e93663c60053f1dd66e9	15.485 ± 0.091	15.421	15.550	1.00
python -m benchmarks.collisions.bench 8d5d889a0f0eb9cfcf1d61122b902b61bfdc7d46	18.247 ± 0.281	18.048	18.445	1.18 ± 0.02
python -m benchmarks.collisions.bench 39633527f4a2d3ab8454a74dd284a89b9ceb2bef	18.573 ± 0.059	18.531	18.615	1.20 ± 0.01
python -m benchmarks.collisions.bench 68a5c838dad6afd5e996ee8ea3a7d2c96d243333	18.911 ± 0.214	18.759	19.063	1.22 ± 0.02
python -m benchmarks.collisions.bench 5fcbcb68600d07b4957a0f797351ba87deb16620	18.868 ± 0.134	18.774	18.963	1.22 ± 0.01
python -m benchmarks.collisions.bench d5cbb312c4c8a83388acfb96f83a683a527d87fc	18.868 ± 0.153	18.760	18.976	1.22 ± 0.01
python -m benchmarks.collisions.bench e1aebe50f5f79908f361ce587736e4b5c3d09001	20.312 ± 0.538	19.932	20.693	1.31 ± 0.04
python -m benchmarks.collisions.bench 89d03fd156376d9c1ff8772ce11d3c062f74c1de	19.919 ± 0.163	19.804	20.034	1.29 ± 0.01

Command	Mean [s]	Min [s]	Max [s]	Relative
python -m benchmarks.collisions.bench d93ba54ace7feb49cd949856a28d4e88977b1265	16.523 ± 0.381	16.253	16.792	1.05 ± 0.02
python -m benchmarks.collisions.bench 055c42413039aa40c458e93663c60053f1dd66e9	15.762 ± 0.047	15.729	15.795	1.00
python -m benchmarks.collisions.bench 8d5d889a0f0eb9cfcf1d61122b902b61bfdc7d46	18.625 ± 0.114	18.544	18.706	1.18 ± 0.01
python -m benchmarks.collisions.bench 39633527f4a2d3ab8454a74dd284a89b9ceb2bef	18.350 ± 0.028	18.330	18.369	1.16 ± 0.00
python -m benchmarks.collisions.bench 68a5c838dad6afd5e996ee8ea3a7d2c96d243333	18.650 ± 0.043	18.620	18.680	1.18 ± 0.00
python -m benchmarks.collisions.bench 5fcbcb68600d07b4957a0f797351ba87deb16620	18.926 ± 0.124	18.839	19.013	1.20 ± 0.01
python -m benchmarks.collisions.bench d5cbb312c4c8a83388acfb96f83a683a527d87fc	18.871 ± 0.149	18.766	18.976	1.20 ± 0.01
python -m benchmarks.collisions.bench e1aebe50f5f79908f361ce587736e4b5c3d09001	19.841 ± 0.107	19.765	19.916	1.26 ± 0.01
python -m benchmarks.collisions.bench 89d03fd156376d9c1ff8772ce11d3c062f74c1de	19.939 ± 0.102	19.867	20.011	1.26 ± 0.01

[cspotcode]

I added some new optimization experiments, I improved the benchmark script, and here's an updated report:

Command	Mean [s]	Min [s]	Max [s]	Relative
50804348 HACK for common hitboxes with parallel sides	24.684 ± 0.353	24.331	25.173	1.01 ± 0.02
0c0d5480 top/bottom/left/right: pass generator to min/max	24.430 ± 0.292	24.231	24.858	1.00 ± 0.01
2e32c8f8 get_adjusted_hit_box: avoid rounding off coords after rotation	24.427 ± 0.137	24.232	24.526	1.00
c9a17ce4 sprite.py: avoid math. attribute access by importing individual items at top of module	27.563 ± 0.201	27.311	27.780	1.13 ± 0.01
c2e8d482 check_for_collision_with_list: micro-opt	27.768 ± 0.318	27.526	28.237	1.14 ± 0.01
d1ca44d0 get_adjusted_hit_box: replace generator expression with list comprehension, even though is immediately converted to a tuple	27.413 ± 0.209	27.170	27.597	1.12 ± 0.01
537b2026 _check_for_collision: optimize broadphase	28.061 ± 0.284	27.782	28.319	1.15 ± 0.01
574c2c92 are_polygons_intersecting: misc tweaks	33.429 ± 0.099	33.342	33.544	1.37 ± 0.01
694acb88 are_polygons_intersecting: remove cast calls	33.726 ± 0.230	33.517	33.967	1.38 ± 0.01
7fb53939 get_adjusted_hit_box: use positional-only params in local function	34.284 ± 0.279	33.868	34.455	1.40 ± 0.01
f76be2f2 get_adjusted_hit_box: inline rotate_point() call	34.454 ± 0.191	34.258	34.646	1.41 ± 0.01
303c5056 get_adjusted_hit_box: avoid self. attr lookup in _adjust_point closure	36.349 ± 0.108	36.271	36.509	1.49 ± 0.01
195461c4 Add benchmark script	36.194 ± 0.281	35.917	36.583	1.48 ± 0.01