test: isolate typeof fixture checks from query setup

[exKAZUu]

Why

The Intel post-merge job at 112bb86 timed out the existing typeof test twice: 10.9s initially and 7.1s on its single failed-job retry against a 5s deadline. One test currently includes loading the grammar, compiling three queries and checking three independent fixtures. Local actual API measurements assign about 169ms of 202ms to initialization, with fixture checks around 22ms, 5ms and 6ms. The CI log cannot attribute the remote time to a specific stage; the setup-heavy lifecycle is established locally, while runner contention remains a hypothesis.

Requirements

• required: Improve all resolvable issues across the requested grammar repositories and prevent duplicate PRs.
• required: Run review-booster and complete-pr simultaneously for each PR, then release.
• required: Fix CI failures after investigating their cause and preserve the existing public parser/query/incremental assertions.
• required: Preserve primary and other owners' checkouts.
• chosen: Separate deterministic Wasm/query initialization from individual fixture checks because the existing Intel typeof test exceeded its combined five-second deadline twice, while local actual API measurements put about 80% of its time in query initialization.
• chosen: Compile the actual shipped queries once in a bounded 30-second setup hook, matching the existing configured-query compilation test's budget, and keep the five-second default for each fixture; no retry, mock, skip, performance-threshold or global-timeout change is justified.
• chosen: Give each of the three fixture tests its own parser and trees because its assertions and restoration are independent; share only the immutable loaded language and read-only compiled queries, and delete all resources in lifecycle cleanup.

Customer Summary

Keep the typeof compatibility and edit checks reliable on slower runners, with a distinct failure for each fixture.

Technical Summary

Move real Wasm/query initialization into a setup hook with the existing query-compilation test's 30s budget. Run the three unchanged fixture bodies as separate tests, each with its own parser/trees and the default 5s deadline. The hook owns the shared read-only queries and releases them after the suite. Grammar, generated parser, Wasm, runtime, dependencies and the error-recovery CPU assertion are unchanged.

Testing

The entire original fixture body, including all domain assertions, actual edits, fresh-tree comparisons, restoration and tree cleanup, is byte-exact. Fresh generation/build passes and eleven production/config/performance sources plus rebuilt Wasm are unchanged. Lint/types and the typeof/conversion/definition/query checks pass four files/15 tests in 4.8s. After integrating merged PR #67, lint/types and the typeof/operator checks also pass two files/four tests in 2.1s. All current-head CI checks pass. Readiness 1791271189-2787e01b8585b312 reports PASS at 5b6a653f79b95c24bd2339a049de05c20f5b8bf9 with no new conversation events. Review 0b8feadf completed with that same last-reviewed/completion head, reviewed by successful exploration sessions: five successes, two unavailable Claude routes, no findings. All three saved notes and seven job contexts were inspected; one reviewer could not run the test without an unbuilt Wasm, while author and CI checks cover execution. No provider retry or new retrospective issue was warranted.

Notes

This is a separate test-lifecycle follow-up to merged PR #66. Published C++ 1.5.5 passed official package, native, browser, Cargo and corpus retention checks; its post-merge Intel job remains failed. No further rerun of that source or relaxed performance assertion is used. Local timing is not a general performance guarantee.