chore: merge upstream pingdotgg/t3code through b72740852f

[andrewcai8]

Merges 222 upstream commits, through upstream's newest commit with green CI. Notable ones: a full Claude context window no longer shows as a usage limit (pingdotgg#19848), Claude recovers after capability probe failures by reusing the last good account once (pingdotgg#21042), and the mobile new task flow, git sheets and thread settings compile under the React Compiler (pingdotgg#21141).

Conflicts:

• Claude driver, provider, usage limits UI, connection supervisor and registry, MCP tool descriptions, bootstrap thread startup: resolved as in the first attempt at chore: merge upstream pingdotgg/t3code through d7c711503f #223, where both sides of these files were the same as now. The fork's host probe now takes upstream's failing probe. When a probe fails and upstream has no last good account to reuse, the probe asks claude auth status. A probe that read no usage is cached for no time, which replaces getProbeDroppingFailedUsage.
• Mobile new task flow, thread list actions, usage limit card, workspace layout, thread screen, scheduled tasks, composer command menu, pending new tasks: took upstream's React Compiler rewrites (no manual useMemo/useCallback) and kept the fork's cloud placement, account switch, settlement scoping and queued cloud messages.

Fork follow-ups the merge needed:

• runClaudeCommand runs with stdin ignored. The fork's claude auth status fallback otherwise waited forever on a CLI that reads stdin, which upstream's probe-recovery test exercises.
• The account switch tests' limited Claude result carries api_error_status: 429. After fix(server): a full Claude context window no longer shows as a usage limit pingdotgg/t3code#19848, blocking_limit alone means a full context window.
• The usage limit card's account switch reads its command result like upstream's toggle does, so the card compiles. The fork's three cloud connection files stay on the compiler's known-bailout list.
• serverRuntimeStartup.test.ts expects the bootstrap thread on the first enabled instance when codex is disabled, which is the fork's autoBootstrapModelSelection. McpHttpServer.test.ts provides the fork's Home services.

Verification: typecheck is clean for contracts, shared, client-runtime, server, web, mobile and desktop. Mobile React Compiler coverage passes (899 tests). The server tests for every file both sides changed pass (26 files, 806 tests).

🤖 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	4.9 KiB	5.0 KiB	+23 B (+0.5%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+23 B (+1.9%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.8 KiB	20.9 KiB	+41 B (+0.2%)	29.3 KiB	✅
Codex	Live turn messages	1	2	+1 (+100.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	−31 B (−0.6%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	−31 B (−2.5%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	−41 B (−0.2%)	29.3 KiB	✅
Claude	Live turn messages	2	1	−1 (−50.0%)	8	✅

Baseline: 6f37f7c · PR result: 8678746 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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