1122 test path smoothing with splines

[mFragaBA]

Motivation

We want to make the bot movement a bit less robot-like. For now we won't mind it not being perfect. We compared against brawl stars and tried to use their bot movement as a baseline

Closes #1122

Summary of changes

Added 2 core functions for bots:

• 1 that calls a rust nif that implements path simplification. This means that it removes points in the path that are not actually needed. In order to do that, we have to do "raycasts" by checking collisions between lines and the game obstacles. This is definitely not efficient enough (as we need to check collisions between each path's edge and all obstacles in the map) and can be improved but is good enough for a first iteration
  • based on ian millington's book on AI programming for games
• another one that takes a path and returns a "smoothed" version using splines. This one was written using pure elixir, we might aim to implement this in the future using rust to improve it's runtime perf. This feature is gated under an env var flag TEST_PATHFINDING_SPLINES. Based on load tests there is a noticeable performance impact of doing this, so for now we'll keep it off by default. Once we retake bots research we can use it to compare against other techniques or try to improve it.
  • based of https://qroph.github.io/2018/07/30/smooth-paths-using-catmull-rom-splines.html

How to test it?

Play a game locally. Check that bots take more reasonable paths. You can test the splines by running export TEST_PATHFINDING_SPLINES=true and restarting the app. Bots will take more "rounded" paths, the movement is still not 100% smooth, but it is done so in small increments.

Checklist

• Tested the changes locally.
• Reviewed the changes on GitHub, line by line.
• This change requires new documentation.
  • Documentation has been added/updated.

[mFragaBA]

Ran loadtest without using splines. Here's the snapshot. While metrics didn't vary that much, the game felt a bit more stuttery compared with main with 400 games, but run smoothly with 300

[mFragaBA]

Just ran another loadtest but with the splines feature on. The games felt more stuttery. We'll merge this PR but keeping the flag off by default with a note on splines. In the future if we double down on them we can look for improvements such as:

• implementing the algorithm with rust
• play around with spline parameters
• try other path smoothing techniques

[mFragaBA]

we need to store the obstacles in order to perform collision checks when simplifying the path

[mFragaBA]

you can also play with these values. I used the values suggested in the article, and played around with the segment_point_amount. After 5 I didn't see a difference