fix(planning): drop the Live grace on a fresh interview page and name the waiting event in the header #8170

Refs: #7967
Depends on: #8169

Refs: #6053

Agent Brief

Type: Bug
Summary: A fresh interview page with no watcher never reads Live, and an event waiting with nobody listening shows in the header, not only as a list chip.

Current behavior:
On a fresh load with no watcher, the header reads Live for the first DEAF_AFTER (30 s), because the not-listening check waits that long before it fires. When an event is waiting and no session is listening, the clearest sign is a per-question chip (No session is listening; type next in the terminal); the header shows only the short Not listening pill, and only after the grace.

Desired behavior:

• No waiter and no lease on a fresh load: the header shows the not-listening state at once (the first-wake text when no watcher has ever polled, else Not listening: type next, or Idle when nothing waits on Claude), with no window in which it reads Live.
• An event waiting with nobody listening (no waiter, no lease, or a delivery unhandled past the stall window) is named in the header, including which question or event waits.
• Live and the lease-held working state behave as the sibling item set them.
• Finished, closed and offline states keep their current text.
• Per-question chips and the receipt line keep today's behavior, including their 30 s grace.

Key interfaces:

• The page's header state function: its not-listening branch applies on load without the DEAF_AFTER grace, and its text names the waiting event.
• DEAF_AFTER: keeps governing the chips and receipt line; the header no longer waits on it.
• /api/state listener fields (waiters, lease, idleFor): read only, unchanged.

Acceptance criteria:

• In the diff, the header's not-listening branch does not depend on DEAF_AFTER elapsing.
• A UI test in the surface's playwright suites covers: no watcher on first load reads not Live within 5 s of load; an event waiting with no watcher shows its id in the header. Add the assertions to an existing suite phase where its server state fits, rather than a new server.
• Existing chip and receipt-line assertions in the surface suites pass unchanged, and any existing header (#pill) assertion the change breaks is updated to the new text in the diff.
• scripts/run-test-file.sh plugins/planning/surface/surface.test.sh passes (browser suites may print SKIP where playwright-cli is absent; CI's surface job runs them).
• A planning changelog fragment is added and scripts/check-changelog-fragments.sh passes.

Out of scope:

• The Live and lease-held working states (sibling item; land it first, as both change the same header function).
• The session bridge (lib/session-bridge/ and its generated copies under plugins/planning/surface/): no transport change; registered in scripts/shared-copies.txt.
• Per-question chip and receipt-line texts and timing.
• Wake batching, auto-handling of accepts, layout and accessibility papercuts Q6-Q9, Q11 (sibling items).
• skills/interview/SKILL.md text.

This was generated by AI during backlog shaping. Split from #7967, whose lane run timed out at 25 minutes on the combined scope.

This was generated by AI during backlog shaping. Split from #7967, whose run used the whole 25-minute budget on the work itself.