feat(planning): auto-handle plain accepts and confirms so they wake no agent #7969

This was generated by AI during attend-queue.

Refs: #6053
Depends on: #7968

Agent Brief

Type: Feature
Summary: The interview server marks a plain accept (no note text) and a plain confirm handled when it records them, so the watcher wakes only for events that need a reply.

Current behavior:
Every saved event stays unhandled until the agent sends a handle op, and the watcher (which waits with after=handled) wakes for any unhandled event. 47 of 54 audit-run events were plain accepts or commitment ticks with no text. The watcher's event selection goes through planning's unhandled() override, so this is planning-owned. Owner decision recorded on #6053 (2026-10-13): server auto-handles plain accept and confirm after testing that sync-ledger and exports still record them.

Desired behavior:

• A plain accept with no text and a plain confirm are recorded in responses exactly as today and marked handled at save time.
• Events that need a reply stay unhandled: ask, own, rephrase, note, research, cancel-research, undo, wrapup, confirm-understanding, accept with text, accept-audit.
• When an auto-handled accept settles the last open question of a round, an event the watcher wakes on is still produced, so the agent learns the round is complete.
• sync-ledger and export-brief record every auto-handled accept, the same as a hand-handled one.
• Undo of an auto-handled accept keeps working: an auto-handled plain accept stays undoable until the agent acts on a later event in that round. The server records auto-handled seqs apart from agent-handled ones, and the undo refusal "Claude already handled it" applies only to agent-handled seqs (decision defaulted on this issue; veto before merge).
• The interview surface reference describes which events are auto-handled; the dedicated-session offer stays.

Key interfaces:

• Planning's Hub.unhandled(r) and the save path: the handled set gains auto-handled seqs.
• Responses and questions documents: unchanged shape.
• sync-ledger and export-brief outputs: unchanged for the same answers.

Acceptance criteria:

• A test written first shows sync-ledger and export-brief output identical for a round answered with plain accepts whether those events were handled by the agent or auto-handled. The PR's CI runs plugins/planning/surface/test_exporters.py and it passes.
• The PR's CI runs plugins/planning/surface/test_server.py and it passes. It includes cases for: a plain accept absent from unhandled(); an accept with text present; the last plain accept of a round yielding a wake-producing event; undo of an auto-handled accept succeeding; undo of an agent-handled accept still returning 409.
• The diff touches no file registered in scripts/shared-copies.txt. Check: the PR's CI shared-copy sync check passes.
• A planning fragment exists (scripts/new-changelog-fragment.sh planning patch), and scripts/check-changelog-fragments.sh --check passes.

Out of scope:

• The bridge transport and watch.sh (shared copies).
• The settle-window setting (sibling item it builds on).
• skills/interview/SKILL.md text.

This was generated by AI during backlog shaping. Split from #6053; the owner approved filing it in an attended backlog-shaping session on 2026-10-18.