Extend cargo-mutants beyond checksum/bcd/party to the full pksave crate

[kasbuunk]

The mutation pass in issue #5 covered only checksum.rs, bcd.rs, party.rs (206 mutants, 5 missed → fixed); the other ~15 core modules (boxes, items, pokemon, stats, text, validate, daycare, hof…) have never been mutation-tested, so their 98% line coverage may still hide weak assertions.

Instructions:

1. Run cargo mutants -p pksave (whole crate; expect ~1h — use --in-place --jobs N or shard by -f per module if the runner is slow).
2. For each missed mutant: add a killing test, or record it as provably-equivalent with a one-line rationale (the |-vs-^-on-disjoint-nibbles precedent from Review: test coverage — llvm-cov, mutation testing, high-value GUI tests #5).
3. Consider a weekly scheduled CI job (cargo mutants --in-diff on PRs is cheap; full runs on cron) — only if the full run is under ~30 min with caching.

Alternatives considered
Leaving the remaining modules to line coverage alone — but #5 showed mutation testing finds assertion gaps 98% coverage misses.
Mutating the app crate too — lower value; its logic is thin over the core.

[kasbuunk]

Full-crate mutation pass complete; every killable survivor killed (tests merged in #44 and #45).

• Run: cargo mutants -p pksave — 1555 mutants in 53 min (2 jobs): 89 missed, 1352 caught, 113 unviable, 1 timeout (ceil_sqrt +=→*= is an infinite loop — detected by timeout, fine).
• Killed (72): dedicated tests for every substantive survivor — box/party layout arithmetic pinned against raw bytes at slot indexes ≥ 1 (symmetric read/write paths can self-cancel, so views alone don't cut it), is_empty polarity through every view and mut wrapper, counts exactly at capacity, level-100 exp coherence, set_bytes at the buffer end and the SRAM/tail boundary, dex bit 151 filtering, exact diagnostic spans (box sentinel, unknown item id, zero quantity), item-list region extents as literals, changed_ranges tail merging, the SaveGame impl via trait object, MonView::moves slot order, flag/map getters. Two verification re-runs over the affected functions confirm: all killable mutants now caught (522 + 88 re-tested). The re-run also caught 5 first-attempt tests whose data made the mutation a fixed point (slot 0 → +0 == -0; index 2 → 2+2 == 2*2) — worth remembering when writing mutation killers.
• Provably equivalent (17): recorded in the new .cargo/mutants.toml — eleven |→^ on disjoint-bit packs (the Review: test coverage — llvm-cov, mutation testing, high-value GUI tests #5 precedent) excluded per function with one-line rationales; six + COUNT-where-COUNT == 0 mutants left visible on purpose, since a function-scoped exclude would also mask future real arithmetic regressions in those functions. Future full runs should report exactly those six as missed.
• Scheduled CI job: not added — the full run takes ~53 min even warm, well over the ~30 min bar the issue set. .cargo/mutants.toml makes ad-hoc full runs one command with a clean expected baseline.

Alternatives considered: excluding the six +0 equivalents too (hides real regressions in the same functions); sharding a weekly CI run (per-module -f shards fit the time bar but add workflow complexity for a crate whose survivors are now documented noise).

---

Generated by Claude Code