[e54-bot]

Summary

Adds one table test per public surface that reports provider or session failures, per the #1022 owner decision:

• crates/wright-driver/tests/check_failures.rs — the check envelope. Every ProviderError variant (injected through an LppSourceProvider over a scripted LanguageProvider) plus the session failure kinds that reach the surface: the CompilerSession::new construction diagnostic (invalid-selection), session-config-changed, and the source-provider-*/profile-workshop-only load refusals. Each row pins the diagnostic code, the class, and the exit code the class resolves to.
• crates/wright-driver/tests/lookup_failures.rs — the tool-service Lookup response. Connect-channel failures (lpp/initialize) and request-channel failures (lpp/lookup) driven by a scripted provider process, plus the session-boundary failure kinds. Each row pins ToolErrorInfo.code and class (or the unavailable verdict). Notes in the file record the failure kinds that cannot reach the surface.
• crates/wright-language/tests/provider_failures.rs — LSP diagnostics on an .opy document. Same connect/request-channel coverage plus the session-construction diagnostics; each row pins the published code and error severity (this surface carries no exit class).

lint-rule-error is deliberately unpinned on the driver surfaces: Diagnostic.class declares SourceError, but the CLI wrappers that can emit the diagnostic — run_configured and run_lookup in crates/wright-cli/src/main.rs and serve::run in crates/wright-cli/src/serve.rs — discard diagnostic.class and hardcode exit USAGE for every session-construction failure. The declared class and the reachable surface behavior disagree; which side is correct is left for a separate owner decision, and this PR makes no production change.

Removed the coverage the tables subsume: the lookup_failure_class unit table, opy_lookup_carries_an_initialize_refusal_with_its_code, opy_lookup_reports_unavailable_when_the_provider_lacks_the_capability, lpp_source_provider_distinguishes_failure_classes, and the refusal/diagnostics halves of two wright-language service tests (the version-negotiation and rename halves remain — they cover mechanisms the tables do not). The only production diff is the deleted mod tests; for the removed unit-test file; no behavior change.

Verification

• cargo test -p wright-driver -p wright-language — all suites pass (the new tables included).
• cargo test -p wright-cli — all suites pass.
• cargo test --workspace --all-targets --all-features — pass.
• cargo clippy --workspace --all-targets --all-features -- -D warnings — clean.
• cargo fmt --all -- --check — clean.
• git diff --check — clean.
• Mutation checks (issue acceptance criteria), each applied, tested, and reverted:
  • provider_failure_code mapping refusals to unavailable → all three tables fail.
  • lookup_failure_class without the invalid-* arm → lookup table fails.
  • connect_opy_provider relabeling a session diagnostic to provider-refusal-<code> → LSP diagnostics table fails.

Unverified

• UI gates (cd ui && npm ci && npm run check && npm test && npm run build && npm run check:size) — not applicable; the change is Rust test-only, ui/ untouched.
• The three scripted-provider surfaces (provider-io request-channel, client session-state errors after a live connect, provider-not-configured for opy on lookup/LSP) have no deterministic trigger — documented in each table's notes and covered at the layer where they are reachable.

[e54-bot]

Review findings addressed in 7045f4b (test-only; no production change):

• Finding 1 — lint-rule-error rows: removed from check_failures.rs and lookup_failures.rs per the owner ruling — removed, not re-pinned. Each table now records why the kind is deliberately unpinned. Surfaced for a separate owner decision: Diagnostic.class declares SourceError for lint-rule-error, but the CLI wrappers that can emit it — run_configured and run_lookup in crates/wright-cli/src/main.rs and serve::run in crates/wright-cli/src/serve.rs — discard diagnostic.class and hardcode exit USAGE for every session-construction failure. The declared class and the reachable surface behavior disagree; which side is correct is left for a separate owner decision.
• Finding 2 — headers: provider_failures.rs now names the real check_failures.rs/lookup_failures.rs tables; both driver headers and their test docs now state that every row drives the boundary that surfaces the diagnostic — CompilerSession::new/ToolService::new for the session-boundary rows, session.check()/the Lookup response end to end for the rest.

Verification: all three tables pass; cargo test --workspace --all-targets --all-features passes; cargo clippy --workspace --all-targets --all-features -- -D warnings, cargo fmt --all -- --check, and git diff --check are clean.

Ready for follow-up review.