Chapter 14½: inside eval/src

Two files, one per question the eval answers.

main.rs: which model is the LLM. It opens with what a run costs, then CASES (each input with the tools a careful person would accept), then the loop: build the Worker's request, send it over Cloudflare's REST API with the token read from 1Password for that run, parse the reply as the Worker does, and score it.
facts.rs: whether Jev's first request reads inputs the way the rules need. 25 cases, each with the facts it should come back with, sent through jev-http so the shared spend ledger admits it.

Both spend real money or neurons, so neither runs in cargo test.

File	What
main.rs	The LLM eval, and the entry point for `lmjtfy-eval`.
facts.rs	The facts eval, `lmjtfy-eval facts`.