A repository's runs page filters by state: `?state=failed` lists only failed runs, under a row of state links #6490

/:owner/:name/runs?state=<s> now lists only runs in that state, under a row of links: All · Failed · Running · Succeeded. The new tests, the neighbouring runs-list tests, tsc and bun scripts/design-lint.ts pass. I didn't run the whole bun test suite.

• Accepted states: queued, running, ok, failed and cancelled. Any other value is ignored and the page shows every run, as before. The row only links four of them. The other two work if typed into the address.
• The link row: the current link carries aria-current="page", and every link stays on the runs page. Choosing a state keeps the ?word= filter and the "next page" link. The "Show Agent runs · Deploys · Checks" links under the list still reset the state.
• Failed agent runs: the default list hides finished agent runs (stopped ones included), so ?state=failed would have dropped them. With a state chosen and no ?word=, agent runs are now kept.
• Visibility: unchanged. An outsider to a private repository gets the same 404 as for a repository that doesn't exist. The test checks this.
• File touched: the page is drawn in src/web/studio/lists.ts, not src/web/pages/runs.ts. That file only forwards to it.
• Tests: test/runs-state-filter.test.ts seeds a failed and a succeeded check run. It checks that ?state=failed lists only the failed one and marks Failed as current, and that ?state=bogus lists both and marks All. It also checks the private-repository 404.

To try it, open /<owner>/<name>/runs?state=failed.

Closes #6477

/:owner/:name/runs?state=<s> now lists only runs in that state, under a row of links: All · Failed · Running · Succeeded. The new tests, the neighbouring runs-list tests, tsc and bun scripts/design-lint.ts pass. I didn't run the whole bun test suite.

• Accepted states: queued, running, ok, failed and cancelled. Any other value is ignored and the page shows every run, as before. The row only links four of them. The other two work if typed into the address.
• The link row: the current link carries aria-current="page", and every link stays on the runs page. Choosing a state keeps the ?word= filter and the "next page" link. The "Show Agent runs · Deploys · Checks" links under the list still reset the state.
• Failed agent runs: the default list hides finished agent runs (stopped ones included), so ?state=failed would have dropped them. With a state chosen and no ?word=, agent runs are now kept.
• Visibility: unchanged. An outsider to a private repository gets the same 404 as for a repository that doesn't exist. The test checks this.
• File touched: the page is drawn in src/web/studio/lists.ts, not src/web/pages/runs.ts. That file only forwards to it.
• Tests: test/runs-state-filter.test.ts seeds a failed and a succeeded check run. It checks that ?state=failed lists only the failed one and marks Failed as current, and that ?state=bogus lists both and marks All. It also checks the private-repository 404.

To try it, open /<owner>/<name>/runs?state=failed.

The runs page now filters by state: ?state= plus a row of state links. I didn't change any code in this round. The only failure was a timing test that has nothing to do with the filter.

• What failed: the readPlan budget test in test/milestones-speed.test.ts took 353 ms against its 30 ms budget. It ran in a shard that was under heavy load, and the log shows three other shards killed on timeouts in the same run.
• Why I think it's load, not this pull: the diff touches only src/web/studio/lists.ts and test/runs-state-filter.test.ts, and readPlan isn't reached from either. Run on its own three times, the readPlan test file passed every time.
• What I checked: test/runs-state-filter.test.ts and test/milestones-speed.test.ts pass together, and tsc --noEmit is clean.
• What I did: pushed an empty commit so the forge re-runs the check on a quieter machine.

If readPlan fails again, the test is flaky under load and needs its own fix. This pull shouldn't carry it.

To try the filter, open a repository's runs page and follow the state links, or add ?state=<state> to the URL.

A repository's runs page now filters by state: ?state=failed lists only failed runs, under a row of state links. The row is now built with seg(), as @juniper asked. bun test test/runs-state-filter.test.ts and tsc --noEmit pass, and the design lint is clean. I did not run the forge's full check; it runs on this push.

• Row: the hand-rolled <nav> with · separators is gone. seg() sits inside a <nav aria-label="Run state">, so the landmark stays and the links get the tap size and aria-current styling from the design system. The word filter and pager still keep the chosen state.
• Test: the existing regexes already matched the seg markup. I added one assertion that the row has class="seg".
• README: I added a line under "What works today" saying how to use the row. I didn't find a state option in the terminal's runs list (I only searched for one), so the README says the filter is web-only for now.
• Cleanup: I removed the now-unused raw import from lists.ts.

The earlier red check was the readPlan timing budget tripping under shard load. It is unrelated to this pull, and I pushed an empty commit to re-run it. If it fails again, that test is flaky under load.

A repository's runs now filter by state on both doors: the web page takes ?state=failed under a row of state links, and the terminal has a new Runs screen with the same filter. test/tui-*.test.ts (439 pass), test/runs-state-filter.test.ts and tsc --noEmit pass. The forge's full check runs on this push.

• Terminal: the terminal had no repository-wide runs list, only the production-deploys screen (deploysScreen). I added a Runs screen, src/tui/screens/runs.ts. In a repository, H opens it and s cycles All, Failed, Running, Succeeded. It reads through listRepoRuns with state, so it is gated like every other run read. I used H because D is already "delete file" in the workspace; my first try on D broke an existing test, and H fixed it.
• Tests:

- Terminal: test/tui-runs-state.test.ts cycles the filter on a repository seeded with a successful run, a failed check and a failed agent run. It also checks the "No running runs" message. - Web: test/runs-state-filter.test.ts now seeds a failed agent run and asserts it shows under ?state=failed. That would fail without the branch that keeps agent runs when a state is chosen. The header now cites #6490.

• README: the line now says both doors filter and names the keys.
• Not done: the terminal list shows the first 50 runs and does not page. The web page pages with a cursor.
• Preview: I haven't looked at the filtered page on the preview. The seeded alice/notes path returns 404 there, so the web filter is covered by tests, not by a look at the preview.