Add RxEvolve example

[albertpod]

Adds a self-contained example demonstrating how to use LLMs to generate nodes and rules, following the discussion. The example includes a brief narrative and a step-by-step walkthrough for building such a system.

Currently slow and relatively expensive: ~$0.07 per call × ~10 calls ≈ $0.70 per run.

WIP — do not merge yet. There are a few open decisions around file generation (e.g., output structure and naming) that should be refined.

Feedback requested

• File generation approach (structure, naming, and reproducibility)
• Any changes you’d suggest to the example or walkthrough

[wouterwln]

I really like this setup and it's impressive it already "works" in a sense it generates at least numerically stable code that produces any inference result. Some thoughts:

• The verbosity of the print statements feels very vibey, which is fine if you want to give that off but wouldn't be my preference. (including emoji's and such)
• "90% success rate within 10 attempts" is of course strange to say because this is 1 attempt and it fails
• The lognormal parameter means bar chart is the wrong way to plot this I think
• the naming convention of the files is fine by me, I don't think it matters that much
• The code generated is very very very shady, I don't know if you want to make a note of it in the example but it's good to know anyway.

I would maybe make the main text a bit more verbose over the code being very verbose. I think the way in which you construct the prompt is cool and maybe that deserves a little bit more attention? All in all a really cool initiative, really curious to see where it goes.