Benchmark with and without shapely

[cspotcode]

I'm creating a PR to ask someone else to run this benchmark, too. Because I'm confused by the results.

I'm running on Windows 11, python 3.11.2:

& 'C:\Program Files\Git\bin\bash.exe' ./benchmarks/bench-shapely.sh

But the important bit is really that it's running benchmarks/collisions/bench.py with shapely enabled and then disabled. I enable and disable using an environment variable. In both cases, shapely is imported -- we pay the startup cost -- but it's only used for collision detection in one of the benchmark cases.

Here are my surprising results:

Command	Mean [s]	Min [s]	Max [s]	Relative
bash ./benchmarks/bench-shapely-helper.sh enabled	90.945 ± 2.481	89.190	92.699	3.54 ± 0.10
bash ./benchmarks/bench-shapely-helper.sh disabled	25.715 ± 0.029	25.694	25.736	1.00

[cspotcode]

Closing to keep the PR queue tidy. This PR is linked in #1536

[cspotcode]

I added a line-of-sight benchmark.

On Windows you can run it like this, assuming hyperfine and git are installed. Or you can run it via bench-shapely-helper.sh directly, or using timeit.

& 'C:\Program Files\Git\bin\bash.exe' .\benchmarks\bench-shapely.sh line-of-sight

Here are my results:

Command	Mean [s]	Min [s]	Max [s]	Relative
bash ./benchmarks/bench-shapely-helper.sh line-of-sight enabled	86.526 ± 1.641	85.366	87.686	1.00
bash ./benchmarks/bench-shapely-helper.sh line-of-sight disabled	89.144 ± 0.628	88.700	89.588	1.03 ± 0.02

[Cleptomania]

We have decided to remove Shapely entirely with #1646. Much of the pure python implementations were faster or only slightly better with shapely, and we are providing better acceleration options for that functionality via arcade-accelerate coming up with Arcade 3.0.