fix(server): reduce retained thread memory in PR lookups and turn startup

[juliusmarminge]

Targets #2829.

PR lookups retained caller thread snapshots through completed cache fibers and lazy diagnostic stacks. A live heap capture attributed about 310 MiB to this ownership chain. The cache now stores lookup results in deferred values and materializes diagnostic stack strings, releasing the caller snapshot while preserving shared lookups, expiry, invalidation, and failure backoff.

Turn startup also loaded history to estimate a handoff budget before discovering that delivery was already complete or deferred. Budget evaluation now happens after those checks.

Verification:

• The isolated GC regression retains the caller snapshot with the original cache and releases it with this implementation, for both successful and failed lookups, while keeping the result cached.
• 114 GitManager tests, six cache behavior/retention tests, and 28 handoff/turn-start tests passed. The 15 cache-related GitManager tests passed again after the final stack-handling change.
• Server typecheck, scoped lint, formatting, and diff checks passed.

These fixes need an updated server build for live comparison. They do not establish that all multi-gigabyte spikes are resolved; required full-history reads remain a separate investigation. No wire contracts or client streaming behavior change.

---

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR changes an existing server PR-lookup cache with new fiber, TTL, invalidation, and retention behavior, and also changes turn-start history loading. It additionally adds a nodeBuiltinImport:off static-analysis suppression in a test, requiring human review under the applicable policy.

^{You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	2	—	8	✅
Claude	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: 5b09254 · 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: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

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