perf: cherry-pick upstream connection, git and checkpoint fixes

[tusharbhardwaj-bk]

Stacked on #221 (fork-marker check): the markers check needs #221's rule to accept cherry-picked upstream lines. This branch includes #221's commit until that merges.

Problem

Nine upstream fixes from upstream/main (e5a46d6) address bkt3's measured slowness: reconnect CPU, event-loop blocking by logs, git process bursts, turn-end checkpoint cost, and the missing desktop keep-alive for running threads. None of them are in the fork yet. Details and measurements: ~/perf-audit-bkt3/10-upstream.md.

Fix

Each commit is applied with git cherry-pick -x in the approved order, with a5da327 moved before 99641fd, which builds on it. The (cherry picked from commit …) trailers let the next upstream merge recognise them.

upstream	what it does	resolution
7931227 (pingdotgg#11811)	no provider refresh on every subscribeServerConfig	clean
b17cc2a (pingdotgg#12305)	bound provider event-log records before serialization	clean
5b377e2 (pingdotgg#11405)	at most 8 short git processes at once	test-file import conflict: kept ours plus the new gitCommandDuration import. The other upstream import there (parseGitCheckoutProgressLine) belongs to a commit not picked.
b12c92f (pingdotgg#10792)	checkpoint capture reuses git index metadata	clean
c1738f1 (pingdotgg#12154)	large sparse checkouts stay on the fast path	VcsProcess.ts: upstream's new onStdoutChunk line sits beside the fork's changed env line; kept both
d17f46d (pingdotgg#12308)	file-search refresh moved off checkpoint processing, coalesced per cwd	source merged clean. In the test file, upstream's new test needs a harness option (secondThreadWorktreePath) from a commit not picked. It is adapted to a single workspace (same assertions: checkpoints finalize while the refresh is blocked, later scans coalesce). Upstream-only test context was dropped.
cb3d95c (pingdotgg#12523)	oversized PR diffs are not retained in the diff cache	added the 4 new lines only (not upstream's unrelated FILES_VIEWED_CACHE_CAPACITY); its 2 new tests are included
a5da327 (pingdotgg#8309)	busy threads keep a fresh disk cache; older-page loading fixed	clean
99641fd (pingdotgg#13554)	desktop keeps running threads synced in the background	import conflict in apps/web/src/state/threads.ts: kept the fork's outbox import plus upstream's OrchestrationThreadShell type

One extra commit, test(web): fork thread fields in the keep-alive fixture, adds the fork-required thread fields to the picked threads.test.ts fixture so it typechecks.

Nothing was dropped. Deliberately not picked, per the approved plan: 5975ec7, eafb4a9, f4600d7, 18de6bb. Step 3 (PR state from GitHub webhooks, #217) replaces that path.

Reaches users: 99641fd and a5da327 are client changes. They reach desktop users only with the next desktop build.

Evidence

Targeted tests (NODE_ENV=test vp test run), all passing:

• GitVcsDriverCore.test.ts + EventNdjsonLogger.test.ts: 89
• GitVcsDriver.test.ts + VcsProcess.test.ts: 49
• CheckpointReactor.test.ts: 36, including the adapted coalescing test
• PullRequestService.test.ts: 121, including both new diff-cache tests
• client-runtime threads-sync + threads-pagination: 39
• web state/threads.test.ts: 3

vp run typecheck is clean in apps/server, packages/client-runtime and apps/web.

Fork-marker check: with #221 it passes. Without it, the 5 upstream-only files fail.

server.test.ts (touched by 7931227) is left to CI; it is too large for the shared box.

Model/harness: Claude Opus 5.5 via Claude Code in T3 Code.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[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	—	13.8 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	57.1 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	10	—	21	✅
Claude	Total thread wire	—	13.8 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	7.3 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket wire	—	6.5 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	57.9 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	9	—	21	✅

Baseline: unavailable · PR result: 5a82338 · 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: 114.9 KiB
• Claude decoded thread snapshot: 115.6 KiB

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