Setup bug bots

[janosh]

coderabbit is configured to require manual invocation which means that bugs usually caught by LLM reviews may slip through.

torch-sim/.coderabbit

Line 4 in ce9ac4c

enabled: false

it's by far the most verbose LLM reviewer so i understand the motivation. https://chatgpt.com/codex, https://cursor.com/bugbot etc. have gotten good and less verbose and likely would have prevented things like the wrong autobatch scaling for mixed PBC. we should setup one or both for https://github.com/TorchSim/torch-sim

[orionarcher]

I think having a automated reviewer that is hyperfocused on bug finding would be valuable. I think it would be significantly less verbose if we provided a prompt to steer it to only comment on logic that is actually wrong, as opposed to just stylistic.