refactor: adopt tests-first Workshop verification

[e54-bot]

Summary

• replace the evidence-specific conformance, corpus, census, and capture model with direct test results and test artifacts
• preserve expected/observed comparisons, catalog and locale identities, immutable source provenance, and client capture metadata
• update real-project fixtures and provenance, retire tracking fields, and route current ADRs through the tests-first model

Verification

• cargo fmt --all
• cargo clippy --workspace --all-targets -- -D warnings
• cargo run -p workshop-rs --bin workshop-catalog-gen -- check
• cargo test --workspace --all-targets
• ablation: removing catalog identity validation made the focused conformance regression fail

Fixes #241

[Teakowa]

Requesting changes for three blocking #241 acceptance-criteria gaps:

1. crates/workshop-rs-cli/src/corpus.rs:207-226 still does not execute the declared reference comparison. The runner validates the expected artifact's metadata, computes a digest for the observed WIR dump, then returns Matched whenever parse/WIR/catalog validation succeeds. A wrong or unrelated oracle can therefore still produce a matched result. This does not preserve an independent expected result or meaningful real-project regression coverage. Required fix: make the expected result participate in an actual comparison (or stop representing the case as a matched reference comparison), and add a negative regression proving that a deliberately divergent expected result cannot pass.

2. #241 explicitly requires the still-open #10 contract to stop directing future work toward an evidence taxonomy/result architecture. Live #10 still requires an evidence model, provenance-aware fixture classes, expectation/evidence sources, and evidence-specific acceptance criteria. Required fix: reconcile #10's durable contract with the tests-first model before #241 is considered complete.

3. The current ADR cleanup is incomplete. For example, Accepted ADR-0007 still defines an "evidence lifecycle", EvidenceRef, "topology evidence", and dataset evidence as current decision language; other current Accepted ADRs still retain generic provenance/evidence wording despite #241 requiring concrete terms such as source attribution, license, pinned revision/version, source mapping, or runtime validation. Required fix: finish the review of all Accepted/current ADRs and current contracts so the surviving guidance no longer teaches the generic evidence/provenance model. Historical text in ADRs that are fully/partially superseded can remain as history when the supersession is unambiguous.

[e54-bot]

Addressed all three requested changes in f11baaf:

• corpus cases with only a pinned oracle identity now report Inconclusive + NotComparable; they cannot report Matched without materialized expected output. Added a regression with a deliberately divergent reference artifact.
• reconciled open issue Establish a tests-first Workshop corpus and seasonal feature census #10 to the tests-first corpus/result contract; source identity alone is explicitly inconclusive.
• completed current ADR/gameplay documentation cleanup, including ADR-0007 source records and current census/client/source terminology; superseded ADR history remains marked and intact.

Local format, clippy, and workspace all-target tests pass.

[Teakowa]

One blocking follow-up remains from finding 3:

• docs/adr/0001-catalog-boundaries.md still defines a generic Provenance contract ("Every catalog entry and locale table carries provenance" grouping source identity, license/review status, and generator identity), and live #10 still says "provenance-linked regression fixtures". #241 explicitly says not to preserve generic provenance abstractions or that exact provenance-linked regression concept; use the concrete continuing concerns instead.

Required fix: replace the remaining current-contract umbrella wording with the actual fields/requirements (source attribution/identity, license/review metadata, pinned revision/hash, generator/reproducibility information as applicable), and remove the prohibited provenance-linked regression wording from #10. Historical superseded ADR text can remain unchanged.

[e54-bot]

Addressed the remaining provenance wording follow-up in e824f3a:

• ADR-0001 now states the concrete source identity/attribution, license/review metadata, generator identity, and pinned revision/version/content-hash requirements instead of a generic provenance umbrella.
• Issue Establish a tests-first Workshop corpus and seasonal feature census #10 now uses source attribution and source-attributed regression fixtures; the prohibited provenance-linked wording is removed.

The branch is clean and pushed at e824f3a.

[Teakowa]

LGTM