fix(threads): keep un-settled threads active after their PR merges

[gfsaaser24]

The bug

Un-settling a thread whose PR has merged only sticks for minutes. Repro (observed on worktree t3code-767f58a0, PR #54 merged):

1. Un-settle the thread → the server stamps the keep-active pin (settledOverride: "active").
2. Open the thread / a session starts → the decider treats this as "real activity" and clears any override back to neutral — including the pin you just set.
3. The session goes idle (or the reaper ends it) → nothing blocks classification anymore, the override is gone, and the merged-PR rule re-settles the thread instantly.

The thread reads as "done" while follow-up work is still happening in its worktree, and every un-settle buys only one burst of work.

The fix (two parts, intended behavior preserved)

1. Sticky keep-active pin (server, decider.ts, 3 sites). Activity still wakes an explicitly settled thread — blocked work must never stay hidden, that direction is untouched. But activity no longer spends the user's explicit "not done" pin. Only an explicit settle clears it.

2. Post-merge-activity gate (client, threadSettled.ts). Instant merged/closed-PR auto-settle now applies only when the thread went quiet at or before the PR's completion. Activity newer than the PR's updatedAt defers to the ordinary inactivity rule. To feed this, the PR's updatedAt (already fetched from gh, previously dropped) now rides VcsStatusChangeRequest into the web and mobile classifications. When the timestamp is absent (older servers), behavior is unchanged.

The common case is untouched: a thread that goes quiet when its PR merges still settles immediately.

Upstream context

• [Feature]: Durable per-thread "never settle" pin — the existing "active" override is cleared by activity, so no thread can be kept permanently active pingdotgg/t3code#5575 describes exactly this gap ("the existing 'active' override is cleared by activity").
• fix(threads): make completed-PR auto-settle configurable pingdotgg/t3code#5643 proposes a settings toggle plus sticky un-settle; fix(threads): make completed-PR auto-settle configurable pingdotgg/t3code#5842/feat: allow disabling auto-settle on merge pingdotgg/t3code#5880/feat(clients): add PR auto-settle toggle pingdotgg/t3code#5141 are toggle-only variants. None are merged. The post-merge-activity gate here fixes the default path without adding a setting, and is complementary to those toggles.
• Registered in SEAM.md as an upstream candidate; prefer upstream's version wholesale if they merge an equivalent.

Tests

• New: decider "keeps a keep-active pin through real activity" (turn start, session start, approval request); effectiveSettled post-merge gate coverage (fresh/stale/quiet/absent-timestamp × merged/closed); GitManager updatedAt forwarding.
• Updated: 2 decider tests asserting the old clear-on-activity behavior; 5 GitManager deep-equality assertions now include the forwarded field.
• Green: server orchestration (26), client-runtime threadSettled (180), contracts (226), web logic suites (148), mobile thread lists (71). GitManager: 67/71 — the 4 failures are pre-existing 12–20s timeouts in cross-repo probing tests, confirmed identical on a clean-HEAD baseline run.
• Typecheck, lint, and format clean across contracts, client-runtime, server, web, mobile.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	11.3 KiB	11.3 KiB	−28 B (−0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	5.4 KiB	5.5 KiB	+7 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	5.9 KiB	5.8 KiB	−35 B (−0.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	49.7 KiB	49.7 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	11.3 KiB	11.3 KiB	+21 B (+0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	5.5 KiB	5.5 KiB	−9 B (−0.2%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	5.8 KiB	5.9 KiB	+30 B (+0.5%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	50.6 KiB	50.6 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 315f8af · PR result: 1ff162a · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 94.6 KiB
• Claude decoded thread snapshot: 95.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.