test(gq): register every .gqt case as its own test in a dedicated corpus crate

[azimafroozeh]

What & why

This PR moves the .gqt corpus and its runner into a dedicated crate, omnigraph-gqt, and registers every case as its own test, case::<file>.gqt, discovered at run time by datatest-stable. The #596 walker was one test, selectable only through OMNIGRAPH_GQ_LOGIC_TESTS, invisible to --list, cargo-nextest, and IDE test views.

	Before	Now
Run the cases matching a name	OMNIGRAPH_GQ_LOGIC_TESTS=issue_563 cargo test -p omnigraph-engine --test gq_logic_tests	cargo test -p omnigraph-gqt issue_563 (add --test gq_logic_tests to skip the unit tests)
List the corpus	not possible (the walker was one test)	cargo test -p omnigraph-gqt --test gq_logic_tests -- --list
One pass/fail line per case in cargo-nextest and IDE test views	not possible	libtest-compatible names, case::<file>.gqt
Bound case concurrency	OMNIGRAPH_GQ_JOBS=<n> cargo test …	cargo test -p omnigraph-gqt -- --test-threads=<n>
Refuse a symlinked or non-UTF-8-named case instead of skipping it silently	not possible	corpus_layout
Read a failing case's row diff	inside the harness error, one escaped line	printed under the case's FAIL line
Add a case	a file under tests/gq_logic_tests/, no Rust change	a file under crates/omnigraph-gqt/cases/, no Rust change

The runner is a library (run_case, list_cases) for a follow-up --show binary. The crate is publish = false, outside default-members and the release build.

Backing issue / RFC

• RFC 0045 (docs/rfcs/0045-gq-logic-tests.md): the per-file test identity it recorded as its upgrade path; the body is updated in place and the 2026-09-03 Decision-log entry names what it supersedes.

Checklist

• Change is focused (the corpus and runner move with per-case registration; the corpus-path pins in the gate script, AGENTS.md, docs/dev/testing.md, and the workflow follow)
• Tests added/updated for behavior changes (the 94 self-tests move as unit tests; new: corpus_layout, the symlink and non-UTF-8 battery rows, budget and panic tests on the single-case runner; 7 cases green)
• Public docs updated if user-facing surface changed (docs/dev/testing.md, AGENTS.md; developer surface only)
• Reviewed against docs/dev/invariants.md — no Hard Invariant weakened, no deny-list item hit (test infrastructure; no engine state touched)

Local verification

• cargo test -p omnigraph-gqt: 94 unit tests and 7 cases green
• cargo test -p omnigraph-gqt --test gq_logic_tests -- --list: 7 case::<file>.gqt lines; with issue_563: 2 run, 5 filtered out
• a symlinked x.gqt in cases/: corpus_layout fails naming it
• python3 scripts/check-fix-regression.py --self-test, scripts/check-agents-md.sh, cargo clippy --workspace --all-targets -- -D warnings, cargo fmt --all --check: clean
• cargo test --workspace --locked --no-fail-fast: 102 binaries green; external_blob_file_policy_rejects_special_files red only under a sandbox that forbids UnixListener::bind

Notes for reviewers

• A name filter matching no case is libtest's green zero-test run (the walker refused it); a bare cargo test at the root skips the crate (-p omnigraph-gqt or --workspace reaches it).
• The Cargo.lock windows-sys and socket2 edge moves are cargo's minimal resolution for the new dev-dependency: HEAD's lock plus cargo update --workspace reproduces this lock byte for byte.
• The fix-regression gate runs from the base branch, so its moved corpus path takes effect after merge; this PR closes no issue. The honest routes (new case, strengthened case, Rust-shape test, bless) pass; .GQT and nested files are refused by corpus_layout, the gate, and --list.
• The crate's unit tests run on every pull request, so a #[test] fn issue_N inside crates/omnigraph-gqt/src/ executes there.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.