feat(settings): add typed occurrence operations

[Teakowa]

Summary

• expose typed authored settings values and structured operation errors
• add source-preserving SettingDefinition::read and write operations over existing occurrences
• resolve effective hero applicability from reviewed exceptional hero evidence while preserving unknown common evidence
• reject wrong kinds, non-finite numbers, unknown enum/list members, and source-list resizing
• document and integration-test typed read/write behavior with unrelated structure preservation

Verification

• cargo fmt --all -- --check
• cargo +1.85.0 clippy --workspace --all-targets --locked -- -D warnings
• cargo test --workspace --all-targets --locked
• cargo run --locked -p workshop-rs --bin workshop-catalog-gen -- check
• git diff --check

Refs #111

[Teakowa]

Post-merge QA audit — changes required

A full audit of the merged #111 implementation found contract regressions and unsafe edit behavior. #111 should not remain completed yet.

Blocker

1. Hero applicability is again inferred from locale/presentation data. hero_setting_applicability() uses GENERATED_HERO_SETTING_NAMES and non-empty localized strings to decide whether a hero/key is applicable. This directly contradicts ADR-0006 and the generated file's own contract: that projection is presentation-only, absence/emptiness is not an applicability fact, and locale label quality must never change applicability. Applicability needs independent reviewed semantic evidence or must remain Unknown.

2. A narrower semantic target can silently edit a broader shared occurrence. applicability() intentionally allows Team -> Hero and TeamAbility -> HeroAbility narrowing. write() then accepts those narrower targets, but concrete_path() only substitutes placeholders that exist in the definition's source path. Calling write() on a common TeamAbility definition with a concrete HeroAbility target therefore edits the team/common setting, affecting multiple heroes, even though the caller supplied one hero target. The same problem exists for common Team definitions queried with Hero. A validated source-edit API must not silently widen the mutation scope.

Major

3. Unknown applicability is treated as write-safe. ensure_target() accepts both Applicable and Unknown. For an evidence-insufficient target, write() therefore succeeds without surfacing that the applicability contract is unresolved. At minimum the edit operation needs an explicit unknown/inconclusive path rather than silently treating it as validated.

4. The required evidenced numeric-clamp example is not backed by a catalog definition. Production catalog projection still constructs Number/Percent domains with NumericBounds::unknown(). The docs create an ad-hoc NumericBounds::new(0, 500) value instead of demonstrating an actual provenance-backed Workshop setting. That does not satisfy Expose typed settings query, validation, and source-preserving edit APIs #111's requested representative evidenced clamp API example and should not be presented as proof that the catalog exposes such evidence.

5. The patch-oriented source-edit requirement remains unproven. write() mutates the SettingsNode tree in place and returns (). It preserves semantic tree structure, but it does not expose a source patch/edit over the original source. Expose typed settings query, validation, and source-preserving edit APIs #111 explicitly says source editing should be patch-oriented where provenance is available and whole-file regeneration must not be the default. This needs either a bounded source-edit contract or an explicit rebaseline of the issue; it should not be silently declared complete.

Goal evidence

There is also no durable downstream WrightKit consumer of definitions_by_id/the new typed operations visible in the org repositories yet; current code search finds the API only inside workshop-rs. That is not sufficient evidence for the Goal's real-consumer dogfood condition.

CI on the PR is green, but these are semantic/API-contract gaps not exercised by the current tests. #111 is being reopened and remains dependent on the corrected #110 contract.