test(ai-proxy): end-to-end coverage for ADR-0030 + drive-by dispatch fix

[ndreno]

Summary

End-to-end coverage for the ADR-0030 implementation stack (PR-1 → PR-3) + a drive-by fix to a bug surfaced by the new tests.

Stacked on PR #76 (routes table). Will rebase down the chain as each lower PR merges.

Drive-by fix

The path-based dispatch added in PR #73 was too strict: it returned 404 for any req.path != /v1/chat/completions, breaking every existing test fixture and any operator-defined operation path. The dispatcher shouldn't constrain the operator's choice of path when only one protocol is on offer.

Now defaults to chat_completion::handle for any path; PR-4 will narrowly add /v1/responses when there's a real second protocol to differentiate. The dead 404 arm of error_response() is dropped, and the unit test from PR-1 that asserted the rejected behavior is replaced with one asserting the dispatcher accepts custom paths today.

Coverage added

Integration (crates/barbacane-test/tests/ai_proxy.rs, +5 tests)

• test_ai_proxy_routes_first_match_wins — wiremock with three route prefixes (/route-claude, /route-gpt, /route-catchall) proves the right upstream URL is hit for each model glob via the actual data plane pipeline.
• test_ai_proxy_400_when_body_omits_model — proves the model_required short-circuit fires end-to-end.
• test_ai_proxy_400_no_route_when_model_does_not_match — proves the no_route response shape ships through the data plane.
• test_ai_proxy_403_model_not_permitted_does_not_reach_upstream — wiremock with .expect(0) proves the upstream is never called when catalog policy denies. Catches the regression where a denied model leaks through.
• test_ai_proxy_403_does_not_fall_through_to_next_route — proves the no-fallthrough rule from ADR-0030 §3 holds in the real pipeline (denied claude-opus-* does NOT reach the catch-all ollama route).

End-to-end (crates/barbacane-test/tests/ai_gateway.rs, +2 tests)

• cel_driven_target_deny_fires_403_not_silent_pass — the load-bearing ADR-0030 §3 subtlety in the actual pipeline. cel writes ai.target=anthropic-tier based on a header; the named target carries deny: ["claude-opus-*"]; a request with model: claude-opus-4-6 hits 403, not silent pass. Mock has .expect(0) so the test proves the upstream is never called even though the target was selected via context.
• cel_driven_target_deny_passes_when_model_does_not_match_deny — positive control for the same spec; non-denied model reaches upstream.

Compilation smoke (tests/fixtures/ai-proxy.yaml, +2 operations)

• /ai/routed/chat/completions — full routes table with allow, deny, and a catch-all.
• /ai/restricted/chat/completions — default_target + catalog deny on a named target.

Vacuum migration UX (docs/rulesets/tests/invalid-ai-proxy-leftover-model.yaml + run-tests.sh)

Regression fixture proving the auto-generated dispatch validator surfaces:

Unknown config field "model" for dispatcher "ai-proxy". Allowed fields: provider, api_key, base_url, timeout, max_tokens, fallback, routes, targets, default_target.

— with the full list of allowed fields including routes — when an operator forgets to delete model: after upgrading from ADR-0024. This is the migration UX promised in PR #75's CHANGELOG.

Known gap (out of scope)

Nested-glob lint coverage for routes[].pattern and allow/deny entries. The auto-generated validator doesn't recurse into nested objects, so vacuum can't catch a regex-syntax pattern at lint time today. Runtime catches it via globset compile error from ensure_compiled_routes (covered by unit tests). This is the generator-recursion improvement we discussed earlier — deserves a separate PR that benefits every plugin's nested schema, not a migration-specific lint.

Test plan

• cargo test -p barbacane-test --test ai_proxy — 9/9 pass (was 4 before this PR; the 5 new integration tests + the 4 pre-existing ones that the PR-1 path-match was breaking)
• cargo test -p barbacane-test --test ai_gateway — 5/5 pass (3 pre-existing + 2 new cel+deny end-to-end)
• cargo test -p barbacane-test --test compilation test_fixture_compiles_ai_proxy — passes with the extended fixture
• cargo test --manifest-path plugins/ai-proxy/Cargo.toml — 67 unit tests pass (1 PR-1 unit test relaxed; counts unchanged)
• bash docs/rulesets/tests/run-tests.sh — 15/15 pass (was 14; +1 for the leftover-model fixture)
• cargo build --target wasm32-unknown-unknown --release — clean
• cargo fmt --all -- --check — clean
• cargo clippy --all-targets — clean on production code
• CI green (modulo pre-existing wasmtime advisory addressed in PR fix(deps): resolve RUSTSEC-2026-0114 (wasmtime patch bump) #74)

Stacking

Branched off PR #76 → which is branched off PR #75 → which is branched off PR #73. Will rebase down the stack as each lower PR merges.