Add handler-level tests for Blade variable rename wiring

[mikebronner]

Summary

Implements #95 — adds the first handler-level tests for the Blade variable rename wiring, closing the gap Holmes flagged in the review of #55 (PR #89): the prepare_blade_var_rename / blade_var_rename_edit handlers in main.rs are only exercised indirectly through pure functions, so a wiring inversion (a bypassed/reversed is_template_variable gate, a dropped unresolved-loop refusal) would slip past the existing suite.

Changes

• src/tests/blade_var_rename_handler.rs (new) — four #[tokio::test] handler tests.
• minimal_backend() helper reuses the established LspService::new(LaravelLanguageServer::new).inner().clone() harness (the same one folio_rename.rs / folio_cursor_containment.rs use). The empty document cache drives the filesystem-fallback path in rename_document_text, reading a tempfile rather than an editor buffer.
• src/tests/mod.rs — registers the new module (alphabetical).

Note on the helper vs. "no Salsa actor"

AC #1 asks for a backend with "no Salsa actor." The repo's only established backend-test harness (test_server() in folio_rename.rs) builds through LspService::new(LaravelLanguageServer::new), which spawns a real (but here inert) actor. The two rename handlers under test never touch Salsa — they read the document and call pure functions — so I followed the existing convention rather than introduce a new disconnected-SalsaHandle production API just to drop an inert thread. Happy to switch if you'd prefer the dedicated constructor.

Acceptance Criteria

• A minimal_backend() test helper in src/tests/blade_var_rename_handler.rs constructs a LaravelLanguageServer with an empty document cache, sufficient for the filesystem-fallback path in rename_document_text to resolve a file:// URI pointing at a tempfile
• A handler test calls prepare_blade_var_rename with a cursor on a @php-only variable (no @foreach/@forelse, no controller binding) and asserts None — catching a bypassed/reversed is_template_variable gate
• A handler test calls prepare_blade_var_rename on a valid file-scoped Blade variable and asserts Some(Range) covering the name (not the $)
• A handler test calls prepare_blade_var_rename inside an unresolved loop region and asserts None — covering the cursor_in_unresolved_loop gate
• A handler test calls blade_var_rename_edit on a file-scoped Blade variable and asserts the WorkspaceEdit rewrites all in-scope occurrences and no occurrences outside scope (a loop that re-binds the same name)
• All new tests are #[tokio::test], use tempfile::TempDir, and pass under cargo test -p laravel-lsp with no network or Laravel project required

Test Plan

• cargo test -p laravel-lsp — lib (1760) and bin (305, incl. the 4 new) targets pass. The 8 tests/integration_tests.rs failures seen locally are pre-existing env-fixture dependencies (test-project/.env is gitignored); CI provisions them via cp test-project/.env.example test-project/.env + composer update.
• cargo fmt --check — clean.
• cargo clippy --all-targets --all-features -- -D warnings — clean (matches CI's warnings-as-errors gate).

Fixes #95

[mr-sherlock-holmes]

✅ Approved

Review Summary

• Reviewed PR #165 — four #[tokio::test] handler-level tests for the Blade variable rename wiring, plus the one-line mod registration. Pure test additions (+186/−0), no production code touched.
• CI green: LSP — test, fmt, clippy passed, so the suite compiles and runs.
• All six acceptance criteria met. Each test was verified not just to compile but to genuinely exercise the real async handler path (read URI → filesystem-fallback in rename_document_text → pure-function gates → result), so a wiring inversion would actually trip it:
  • AC #1 — minimal_backend() added; empty document cache; the Some(Range) accept test proves the filesystem-fallback path resolves an unopened file:// tempfile off disk. Note on the "no Salsa actor" clause: the only constructor (LaravelLanguageServer::new) spawns a Salsa actor unconditionally, so a literally actor-less backend isn't constructible. The helper does the right and idiomatic thing — the actor is inert (the rename handlers never message it), exactly as the existing folio_rename.rs::test_server() harness does. The AC's substance (a backend sufficient for the fallback path, not dependent on Salsa) is satisfied; "no Salsa actor" is a benign wording imprecision, and the test comment calls it out honestly.
  • AC #2 — @php-only $secret → None. Not vacuous: bypassing the is_template_variable gate would let variable_spans find $secret on that line and return Some, failing the test.
  • AC #3 — file-scoped $name → Some(Range) covering "name", with the char before the range asserted to be $. Sigil exclusion proven.
  • AC #4 — cursor in the opaque @foreach ($users /* :) */ as $user) → None. The fixture genuinely produces an unresolved loop (see follow-up below), and removing the cursor_in_unresolved_loop gate would return Some.
  • AC #5 — blade_var_rename_edit on file-scoped $item → WorkspaceEdit whose changes rewrite only lines 0 and 4; the loop-shadowed $item on line 2 is correctly skipped (lines == [0,4]). Scope shadowing proven.
  • AC #6 — all four are #[tokio::test], use tempfile::TempDir, no network, no Laravel project.

What's Good

• Excellent doc comments: each test states the exact wiring inversion it guards against, so the why survives. This is precisely the handler-level coverage the original follow-up (#95, from PR #89's review) asked for.
• Cursor positions are carefully placed and self-documenting; the sigil-exclusion and [0,4] scope assertions are tight and non-tautological.
• Reuses the established handler-test harness pattern (LspService::new + inner().clone()) rather than inventing a parallel one.

📋 Non-blocking follow-ups

• matching_paren (laravel-lsp/src/blade_loops.rs:159) is quote-aware but comment-unaware — a ) inside a /* */ comment in a @foreach header desyncs the paren scan, so an otherwise-resolvable loop is treated as unresolved and rename is refused inside it. Safe/conservative today (refuses rather than mis-renaming), and this PR legitimately exploits it to build the unresolved-loop fixture — but it suppresses a valid rename for that (rare) construct. Outside this PR's diff and not named by any AC. Tracking separately.

Ready for @mikebronner to merge.