chore: hygiene sweep for tracked runtime files and dead scripts

[VedantMadane]

Summary

Addresses the safer parts of #57:

• Untrack runtime/personal files: polyfish-rs/saved_state.json, cache/projects.json, *.css.bak / game_old.css
• Delete dead one-shots: resolve_conflicts.py, pull.sh
• Remove unused original_mcts_zero.rs (no mod / callers)
• Replace Vite template polyfish-ui/README.md with a short project blurb
• Expand .gitignore so these stay untracked

Left out of scope (larger/behavior-touching): root package.json dep rewrite, opening-book forcing call-site retirement, broader MCTS dead-code moves.

Test plan

• Inspected git status / removed paths; no module references original_mcts_zero

Closes #57

[kadenstaker]

Reviewed this. Nine of the eleven deletions are clean - original_mcts_zero.rs isn't declared in ai/mod.rs so the build is unaffected, pull.sh / resolve_conflicts.py have no callers and aren't referenced by CI or scripts/check_cli_contract.py, the four game.css.bak / game_old.css blobs are 0 bytes and unlinked from any index.html, and the new polyfish-ui/README.md claim matches main.rs:187-195.

The one thing I'd change before merge is polyfish-rs/saved_state.json. It's classified here as a runtime/personal artifact, but it's a checked-in test fixture with two live consumers.

1. polyfish-rs/tests/elo_test.rs:11 breaks. The test resolves the fixture off CARGO_MANIFEST_DIR, so it's specifically this tracked file. Reproduced by moving the file aside on the base commit:

thread 'test_mcts_matches_ruin_capture_sequence' panicked at tests/elo_test.rs:12:52:
Failed to load saved_state.json: Os { code: 2, kind: NotFound }
test result: FAILED. 0 passed; 1 failed

#[ignore] keeps CI green, so this only surfaces when someone runs the command the test's own comment documents. The PR doesn't delete elo_test.rs, so it stays in the tree pointing at a file that's gone. The new .gitignore entry (plus the pre-existing polyfish-rs/*.json at line 57) also means the fixture can't be restored without git add -f.

2. POST /api/load regresses into a panicking handler. main.rs:967-972 does read_to_string("saved_state.json").expect(...) followed by an .expect() on the parse, and there's no CatchPanicLayer anywhere in src/. The path is CWD-relative and run-server.sh cds into polyfish-rs/, so it reads exactly the deleted file. Narrower than it sounds: save_game writes the same path, so anyone who has ever hit Save is fine - the broken window is a fresh clone where Load is clicked before Save, which was previously impossible because the file shipped. Startup (main.rs:58-68) degrades gracefully; only this endpoint is affected.

Two ways out. Either keep it tracked with a !polyfish-rs/saved_state.json negation after the ignore rules, or commit to the deletion and also drop elo_test.rs plus make load_game return {"status":"error"} on a missing file. I'd take the first - the .expect() in a request handler is worth fixing either way, but that turns a hygiene PR into a behavior change.

Minor, could ride along in the same push:

• .gitignore:76 - cache/ has no leading slash, so it matches a cache directory at any depth, not just the root one that held cache/projects.json. /cache/ scopes it correctly.
• .gitignore:77 - polyfish-rs/saved_state.json is already fully covered by polyfish-rs/*.json at line 57. Worth removing so nobody later assumes deleting just this line is what restores the fixture.
• CLAUDE.md:89 and README.md:53 still describe original_mcts_zero.rs as live, and CLAUDE.md tells contributors to check whether a change needs to land in it. Since that's the agent instruction file it's worth updating here. (expert_pipeline_audit.md:717 mentions it too, but as a historical note, so that one can stay.)

[kadenstaker]

Thanks for this - most of the sweep is good and I'd like to take it. Nine of the eleven deletions verify clean: original_mcts_zero.rs isn't declared in ai/mod.rs, pull.sh and resolve_conflicts.py have no callers in CI or the CLI-contract check, the four CSS blobs are 0 bytes and unlinked, and the README update matches what main.rs actually serves.

Blocking on one file: polyfish-rs/saved_state.json is a checked-in test fixture rather than a runtime artifact. Deleting it fails tests/elo_test.rs (reproduced - it resolves the path off CARGO_MANIFEST_DIR), and it regresses POST /api/load into a panicking handler, since main.rs:970 does read_to_string(...).expect(...) with no CatchPanicLayer in the router. Neither shows up in CI, which is why I'd rather catch it here.

Simplest fix is to keep it tracked with a !polyfish-rs/saved_state.json negation and drop the now-redundant ignore line. Details plus two minor .gitignore / stale-docs notes are in my comment above.