fix(workshop): converge on released workshop-rs 0.1.9

[Teakowa]

Summary

• pin the Wright workspace to released workshop-rs 0.1.9
• keep the canonical catalog-owned raw Workshop check/lint path
• add durable Workshop contract and real-project integration coverage
• run the production wright check / wright lint commands against the owner-pinned corpus

Closes #191

Validation

• cargo fmt --all -- --check
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings
• cargo test --locked --workspace --all-targets --all-features (595 passed, 2 ignored)
• v0.1.9 exact-corpus provider contract validation passed
• v0.1.9 exact-corpus CLI check/lint integration passed for ai-pve, bastion, defend, illari, and rework
• CI YAML parse passed

The defend fixture retains its expected legacy opaque-action diagnostic.

[Teakowa]

Maintainer direction: keep the dependency convergence and real-project coverage, but do not encode temporary acceptance phases as permanent test architecture.

Please rework the new coverage so it protects durable capabilities rather than P0, dogfood, or #191 terminology:

• rename crates/wright-cli/tests/workshop_p0.rs to a capability-oriented integration test such as workshop_real_projects.rs or workshop_integration.rs;
• rename the CI job from Raw Workshop product dogfood to a durable capability name such as Workshop real-project integration;
• remove new Wright-side P0 / dogfood terminology from test names, output prefixes, comments, and environment variable names; use generic Workshop corpus/integration naming instead;
• keep the same production wright check / wright lint execution against the real-project corpus; this is a naming/ownership/longevity correction, not a request to weaken coverage;
• keep Revalidate raw Workshop check and lint after canonical language-surface completion #191 as the reason this durable integration suite is being established now, not as the identity of the suite itself.

workshop_rs::p0 is upstream naming debt and does not need to be refactored inside this PR; consume it as needed for now. A separate workshop-rs follow-up will own that terminology/API cleanup.

The workshop-rs 0.1.9 dependency update and the required LocalizedString exhaustiveness fixes remain in scope and should stay.

[Teakowa]

Follow-up review on the rewritten head: the phase-specific naming correction is now substantially right. workshop_real_projects.rs, workshop_contract.rs, WRIGHTKIT_WORKSHOP_CORPUS_DIR, Workshop real-project integration, and WORKSHOP_CORPUS are durable capability-oriented names. The remaining workshop_rs::p0::* references are acceptable temporary upstream API consumption and are tracked by workshop-rs#96.

Two cleanup items remain before I would treat this as fully converged:

1. workshop_real_projects.rs still asserts Some(5) for the default lint rule count. Do not freeze the current rule inventory as an acceptance number. Assert the durable contract instead (for example, compare against the authoritative enabled/default rule registry if that exact-set contract matters, or otherwise prove the lint path/rules executed without a hard-coded count).
2. crates/wright-driver/tests/manifest_context.rs no longer tests a manifest-backed context after this PR removes the OPY manifest from raw Workshop parsing. Rename the file/test-surface to the capability it now protects (for example workshop_signature_context.rs) so the old ownership model does not remain encoded in the repository.

Keep the current workshop-rs 0.1.9 convergence, canonical catalog context change, real-project CLI integration, and LocalizedString handling.