refactor(server): thread search and git project lookup read V2

[juliusmarminge]

Stacked on #14012.

After #14012, ProjectionSnapshotQuery had two readers left, thread search and GitManager's per-project settings lookup. The lookup was also a bug (Bug 3 in the design): GitManager resolved a thread's project from the V1 projection_threads table, so a V2-native thread never matched. Per-project git settings (commit/PR writer model, writing style, worktree submodules) then silently fell back to the environment defaults for every thread created since V2.

What changed

• Thread search reads V2 only. A new ThreadSearch service (orchestration-v2/ThreadSearch.ts) runs the same ranked query over the V2 projection: one best match per active thread in an active project, user messages over assistant ones, then newest first. It keeps the snippet rules, escaped LIKE and limit. It no longer searches V1 transcripts that haven't been imported yet (the approved design puts V1 search out of scope). ws searchThreads and the MCP t3_thread_search tool use it.
• GitManager's project lookup reads V2. A thread's project comes from ProjectionStoreV2.getThreadShell, and a bare cwd resolves through ProjectStoreV2.findActiveByWorkspaceRoot. Both are required services now, not an optional snapshot query. server.ts provides them to the GitManager layer.
• Deleted, with no readers left:
  • ProjectionSnapshotQuery (service, layer, and the tests of its search and enrichment)
  • ProjectionPipeline (service, layer, test)
  • orchestration/runtimeLayer.ts; OrchestrationEventInfrastructureLayerLive moved to orchestration-v2/runtimeLayer.ts
  • ThreadBackgroundLiveness and ThreadPlanProgress, registries only the snapshot query read
  • the repositories only the pipeline wrote: pending approvals, projection state, thread sessions, turns
  • now-dead helpers isPersistenceError, the exported parseAttachmentIdFromRelativePath, and the exports of MONITOR_TASK_TYPES / INERT_TASK_TYPES

No wire or contract schema change.

Verification

• vp exec tsc --noEmit -p . in apps/server and packages/contracts: 0 error TS / warning TS.
• vpr knip:check: clean.
• vp lint on touched files: no new warnings.
• vp test run, in small groups, all passing:
  • new ThreadSearch.test.ts (real SQLite): user over assistant, newest user wins, streaming/system/archived/deleted threads and deleted projects excluded, limit, literal %.
  • GitManager.test.ts (115), including a new case: a V2-native thread (created only through ProjectionStoreV2, cwd outside the project root) gets its project's textGenerationModelSelection override. It fails against the old V1 lookup and passes now.
  • McpHttpServer.test.ts, worktree/registration.test.ts, PullRequestSyncReactor.test.ts, ws.test.ts, serverRuntimeStartup.test.ts, runtimeLayer.test.ts, ProjectionRepositories.test.ts, integration/transferBudgetV2.integration.test.ts
  • importer and migrations: LegacyV1Cutover.integration.test.ts, LegacyV1ThreadImporter.test.ts, V1ImportBoundary.test.ts, ProjectSettingsUpgrade.integration.test.ts, 055_* (3 files), 056_*
  • replay: OrchestratorReplayFixtures.integration.test.ts + .contract.test.ts, OrchestratorReplayRecovery.integration.test.ts
• Not run: the repo-wide suite, and a real client (no client changes).

Model: Claude Opus 5.5 (Claude Code)

🤖 Generated with Claude Code

---

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This is a large production orchestration refactor that replaces legacy projection reads and infrastructure with V2 services, including a deliberate change to search visibility and GitManager settings resolution. The breadth of deleted shared infrastructure and changed runtime wiring warrants human review.

No code changes detected at 5095efd. Prior analysis still applies.

^{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	—	1.1 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	1	—	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: 5095efd · 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.

[macroscopeapp]

These newly added service-layer imports use named Live bindings rather than module namespaces. Could you import each service module as a namespace and reference its layer through that namespace in OrchestrationEventInfrastructureLayerLive? This keeps service wiring consistent with the Effect service boundary convention.

^{Posted via Macroscope — Effect Service Conventions}

[macroscopeapp]

Correction to my Effect Service Conventions inline comment on apps/server/src/orchestration-v2/runtimeLayer.ts: the two Live imports come from legacy implementation-layer modules, rather than canonical service modules. The namespace convention does not apply to them. Please disregard that finding.

^{Posted via Macroscope — Effect Service Conventions}