test: integration coverage for Folio route find-references usages leg

[mikebronner]

Summary

Implements #97 — adds the explicit end-to-end assertion for the Folio route find-references usages leg (AC #4) that round-2 review of #57 (PR #91) deferred. The wiring was already correct and covered implicitly by the shared reference machinery; this test makes the call-site leg explicit.

Changes

• New unit test folio_route_usages_are_returned_by_symbol_index in laravel-lsp/src/folio_discovery/tests.rs.
• Anchors the expected route name through the real Folio pipeline (inject_folio_routes + folio_name_for_file) rather than hard-coding it, then builds a route('users.show') call-site as a ParsedPatternsData.route_refs entry and feeds it to a fresh SymbolIndex via insert_file.
• Asserts SymbolIndex::find(Route("users.show")) returns exactly that call-site (file path, line, column), and that an unrelated route name returns empty (no cross-contamination).
• No Backend, SalsaActor, or tokio runtime — matches the integration suite's no-async-harness convention.

Acceptance Criteria

• New test in folio_discovery/tests.rs builds ParsedPatternsData with a route_refs call-site for a Folio-injected route name, inserts via insert_file, and asserts find(Route("users.show")) returns the call-site location (file/line/column).
• Same test asserts find(Route("other.route")) is empty (no cross-contamination).
• Test does NOT instantiate Backend, SalsaActor, or any tokio runtime — uses inject_folio_routes to anchor the name plus SymbolIndex + ParsedPatternsData built directly.
• cargo test -p laravel-lsp passes green with the new test (no existing tests regressed).

Test Plan

• New test passes: cargo test -p laravel-lsp folio_route_usages_are_returned_by_symbol_index.
• Full unit suite green: 1724 + 261 tests pass, 0 failures.
• cargo fmt --check and cargo clippy --tests clean.
• Note: 8 tests/integration_tests.rs cases fail in a fresh clone (missing gitignored .env / test-project fixtures) — pre-existing, confirmed identical on the untouched base commit; unrelated to this change.

Fixes #97

[mr-sherlock-holmes]

✅ Approved

Review Summary

• Reviewed the test-only PR for #97 (+71/−0, one new unit test in laravel-lsp/src/folio_discovery/tests.rs). This closes the find-references usages-leg coverage gap (AC #4) that round-2/3 review of #57 (PR #91) deferred into #97.
• All four acceptance criteria met — verified leg by leg, not on faith:
  • AC #1 ✅ — folio_route_usages_are_returned_by_symbol_index builds a ParsedPatternsData with a route_refs call-site entry, feeds it via index.insert_file(&call_site, &patterns) (tests.rs:499), and asserts find(Route("users.show")) returns exactly that location — file_path, line == 17, column == 23 (tests.rs:504-507).
  • AC #2 ✅ — find(Route("other.route")) asserted empty (tests.rs:511-516), a genuine negative control.
  • AC #3 ✅ — no Backend, SalsaActor, or tokio runtime; those words appear only in the doc comment. Construction is RouteIndex::new() / ParsedPatternsData::default() / SymbolIndex::default().
  • AC #4 ✅ — CI green (LSP test/fmt/clippy, wasm check, CodeQL all pass); API signatures all match the tree.
• The test is honest, not vacuous. SymbolIndex::insert_file keys refs under (SymbolKind::Route, name) and find is a real HashMap lookup (symbol_index.rs:120-127, :315-321) — not an echo of the input. The non-tautology pivot is assert_eq!(route_name, "users.show") (tests.rs:484): the name is recovered through the real Folio pipeline (inject_folio_routes → folio_name_for_file), and that checkpoint fails if the pipeline ever produces the wrong name. The file/line/column assertions are meaningful guards against mis-wiring.
• mergeable: MERGEABLE; the BLOCKED merge state is just this review gate.

What's Good

• Anchoring the expected route name through the real pipeline rather than hard-coding it is the right call — it makes the round-trip assertion mean something. 👍
• Clean negative control (other.route → empty) rules out bucket cross-contamination / fuzzy matching.
• Respects the suite's no-async-harness convention exactly as the AC required — synthetic in-memory PathBuf call-site is never read from disk, real FS access is TempDir-scoped, no new deps, no unsafe.

📋 Non-blocking follow-ups

• None.

Accepted limitation (not a follow-up): this test covers the component round-trip (Folio name → SymbolIndex insert/find), not the async-actor find-references handler that wires them together — a handler bug passing the wrong name (e.g. uri vs name) wouldn't be caught here. That's a deliberate, documented deferral: the integration suite intentionally avoids the Backend/SalsaActor/tokio harness ("not economical to fixture"), and this PR's AC explicitly required avoiding it. Tracking it as a new issue would just recreate the loop that produced #97, so I'm recording it here rather than spinning one out.

Ready for @mikebronner to merge.