One function names a pull's check for a person: green, red, running or none #6080

This adds checkWord(run, headOid) in src/apps/check_word.ts, which names a pull's check green, red, running or none. The six new tests pass, but the type check did not run: tsc stopped at "Cannot find type definition file for 'bun'", and I did not run bun install. The forge's check will run it on this push.

• What it returns: green for an ok run of that head, red for a failed one, and running for queued or running. It returns none for no run, a cancelled run, or a run of any other head.
• Inputs: it takes plain data, the run's state and commitOid (or undefined) plus the head oid, and returns a plain string.
• Tests: test/check-word.test.ts covers each case, including a run of an older head. That case checks every state and gets none each time.

I did not wire checkWord into any page or the MCP, so nothing calls it yet. verdict in src/web/studio/review_page.ts still does its own head check and is the obvious first caller.

Closes #6071

This adds checkWord(run, headOid) in src/apps/check_word.ts, which names a pull's check green, red, running or none. The six new tests pass, but the type check did not run: tsc stopped at "Cannot find type definition file for 'bun'", and I did not run bun install. The forge's check will run it on this push.

• What it returns: green for an ok run of that head, red for a failed one, and running for queued or running. It returns none for no run, a cancelled run, or a run of any other head.
• Inputs: it takes plain data, the run's state and commitOid (or undefined) plus the head oid, and returns a plain string.
• Tests: test/check-word.test.ts covers each case, including a run of an older head. That case checks every state and gets none each time.

I did not wire checkWord into any page or the MCP, so nothing calls it yet. verdict in src/web/studio/review_page.ts still does its own head check and is the obvious first caller.

checkWord now has a caller. verdict in src/web/studio/review_page.ts uses it in place of its own head check, so the /review rows name their check through it. The 16 tests in review-page and check-word pass. With the src/ change stashed, the review-row test fails. I did not run the type check or bun scripts/design-lint.ts after this change; the earlier design-lint run (after the wiring, before the test tweak) was clean, and the forge's check runs the rest.

• Visible change: a cancelled run now shows "No check" instead of "Cancelled", as #6071 specifies. I changed the existing review-row test's cancelled expectation to match; the old expectation was the behaviour #6071 retires.
• Test through the page: "Stale pull" now has a running run of an older head, and the test expects "No check" and no "Running". The cancelled case is the part that fails without the wiring.
• Queued rows: checkWord folds queued into running, so a queued run on a /review row now reads "Running" where it used to read "Queued". That is the word set #6071 asks for.
• Relation to checkWords in src/checks/summary.ts: they are different layers, so I left both. checkWord is the verdict: green, red, running or none for a given head. checkWords is the sentence for a run already chosen, such as "Check passed in 45s", with timing and the failing test name. The pull page and TUI call checkWords only for a run they have already picked. They do not yet check that the run belongs to the current head, which is where "check running" can show when nothing is running. I did not touch them, to keep this pull small. The follow-up is to gate checkWords behind checkWord !== "none" on both callers.