Improvements to proof-loading speed on reruns

[ehildenb]

These are pulled out of the CSE PR: #284. This includes:

• Telling our tools to ignore various generated files in the Foundry test directory (ignore Python files from Foundry in our formatting, and ignore generated files in our gitignore).
• Avoid loading lots of versions of a given proof when only one is called for. This is done by changing Foundry.proofs_with_test(...) -> list[Proof] to Foundry.proof_ids_with_test(...) -> list[str], and allowing the call-site to decide whether to load the proof or not. This significantly reduces startup time when there are several versions of a given proof present.
• Renames internal function run_prover to prefixed version _run_prover to avoid confusion with run_prover imported from KEVM.
• Avoids setting up kore-server for already passed proofs, which speeds up the re-run time on a batch of proofs where some proofs have passed and others haven't.
• Removes a little dead-code.