feat(catalog)!: retire non-native ids forcePlayerHero, stopForcingHero, forceThrottle, stopChasingVariable, isFiringSecondaryFire

[e54-bot]

Summary

Keep one catalog identity per native Workshop element (Fixes #297).

• Remove the non-native duplicate ids forcePlayerHero, stopForcingHero, forceThrottle, stopChasingVariable, isFiringSecondaryFire, and the extra Set Allowed Heroes spelling of setAllowedHeroes. Their zh-CN spellings were already carried by the native entries, so none moved.
• Commit the Workshop.codes wiki inventory (227 actions, 264 values). workshop-catalog-gen check now fails when an action or value has an en-US spelling outside it (case-insensitive).
• Update the digest pin, corpus manifest, source-attribution and language-support docs, and tests that named retired ids.

Notes

• The wiki covers en-US only and no independent source for other locales was found, so no locale spelling from the retired entries was kept.
• The corpus manifest was edited by hand: regenerating against the local export drifted from the pinned commit.
• Breaking change to the Rust API (retired ids); consumers (opy-rs) adopt in their own PRs.

Verification

cargo test --workspace, cargo clippy --workspace --all-targets, workshop-catalog-gen check.

[Teakowa]

Three actionable findings:

1. tools/corpus/zh-cn-corpus.json — the PR notes say this committed corpus manifest was edited by hand because regenerating from the local export drifted from the pinned commit. AGENTS.md explicitly forbids manually refreshing committed test/result files and only permits committed fixtures/manifests when they are reproducible, machine-validated inputs. The corpus tests now validate the hand-edited result rather than demonstrating that the declared source/generator reproduces it. Regenerate from the exact pinned source, or otherwise make this update mechanically reproducible and validated before merge.

2. crates/workshop-rs/src/bin/workshop-catalog-gen.rs::native_spellings — the new guard does not fully enforce #297's “one native identity” acceptance criterion. Catalog::load rejects duplicate aliases case-sensitively, while native_spellings deliberately normalizes wiki names to lowercase. Existing names already demonstrate real case-only differences (Move Player to Team vs Move Player To Team, Number Of Players vs Number of Players). A second entry using the alternate casing would therefore pass both checks while representing the same native Workshop element. Reject duplicate action/value native names under the same normalization used for wiki membership, and add a regression test.

3. Release notes — #297 requires the retired public identities to be called out in release notes. This repo creates releases with gh release create --generate-notes, while this PR's title only says keep one catalog identity per native Workshop element; the five retired Rust API ids are only listed in the PR body. Ensure the release-note source that reaches the generated GitHub Release explicitly names forcePlayerHero, stopForcingHero, forceThrottle, stopChasingVariable, and isFiringSecondaryFire.

[e54-bot]

Addressed in the latest push: (2) case-insensitive duplicate detection plus regression test; (3) the PR title now names the five retired ids, so the generated release notes list them.

(1) Not resolved: the pinned export (workshop-data @ d854bf0) is not available to me — the local copy is not a git checkout and regenerating from it drifts (extra locale spellings in locales.json/catalog). The manifest edit only removes the retired entries and adjusts coverage counts. Please regenerate from the exact pinned export before merge, or tell me where to obtain it.

[Teakowa]

Findings 2 and 3 are resolved. One blocker remains:

• tools/corpus/zh-cn-corpus.json: the author confirms this manifest is still hand-edited because the exact pinned workshop-data export is unavailable locally. That still conflicts with the repository contract requiring committed test/result data to be reproducible and machine-validated. Since the corpus tests consume this file, green CI only validates the edited snapshot, not that the declared pinned source/generator reproduces it. Regenerate it from the exact pinned export (or otherwise make the committed update mechanically reproducible and validate that path) before merge.

[e54-bot]

Manifest is now regenerated, not hand-edited: workshop-catalog-gen corpus against the pinned export (workshop-data meta.commit d854bf01, which the local copy does carry — my earlier claim that it was unavailable was wrong; the drift I saw was in locales.json, which also drifts on unmodified main and is left untouched). The output equals the earlier edit except generatorVersion (0.4.2 → 0.7.1).

Caveat: the same run also merges export-derived (OverPy-sourced) locale spellings into the native startForcingHero/stopForcingCurrentHero/startForcingThrottle/stopChasingGlobalVariable catalog entries. I did not commit that catalog change, since #297 says the export is not evidence for keeping spellings; so re-running corpus on this branch is not catalog-idempotent for those four entries. Decision for the owner if that should change.

[Teakowa]

LGTM