chore(upstream): adopt Claude Agent SDK 0.3.260 and abandoned-turn failure reasons (#9135)

[rynfar]

What this adopts

Upstream #9135 (560afffdea), cherry-picked with -x: Claude Agent SDK 0.3.170 → 0.3.260, plus failure classification for results the CLI gives up on. terminal_reason values (api_error, budget_exhausted, prompt_too_long, model_error, and the rest) and 529 overloads now mark the turn failed with a user-facing reason instead of a success-shaped result with an empty error list. It also handles three new system subtypes (control_request_progress, worker_shutting_down, informational) and the model_refusal_no_fallback notice.

This is the reliability half that upstream had and Pylon did not: after #264–#267, a turn can no longer vanish silently, and with this a turn the CLI abandons no longer looks like a clean finish either.

Review context

Targeted upstream review, scoped to the paths involved in the 2026-09-04 lost-message incident. Recorded in .agents/upstream-review.md as a targeted batch; the reviewed-through cursor is deliberately not advanced. The review confirmed upstream has no fix for the incident itself (its adapter and synthetic-turn code are identical, and it has no strict admission gate), and identified two further candidates that are not standalone and are left pending in the ledger:

• #9653 (OpenCode approvals and stop, turn.aborted terminal mapping) sits on four OpenCode commits Pylon has not taken. Adopt as a series in the next OpenCode batch.
• #9167 (continue active threads across server self-updates) rewrites startup reconciliation Pylon reworked for Prime. Manual port.

Verification

• Clean cherry-pick, lockfile included. vp i resolved the new SDK.
• vp test run on ClaudeAdapter.test.ts: 88 passed (85 existing including fix(claude): between-turn status and resume handshake no longer touch the turn lifecycle #265's two, plus upstream's three).
• vp run -F t3 typecheck: clean. vp fmt --check and vp lint on the adapter files and the ledger: clean.

Model: Claude Fable 5.1. Harness: Pylon (Claude Code provider).

---

^{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.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.6 KiB	13.6 KiB	+3 B (+0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+4 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.6 KiB	−1 B (−0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.5 KiB	56.5 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	8	8	0 (0.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.6 KiB	−96 B (−0.7%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+5 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.6 KiB	−101 B (−1.5%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.1 KiB	57.3 KiB	−840 B (−1.4%)	66.4 KiB	✅
Claude	Live turn messages	8	8	0 (0.0%)	21	✅

Baseline: f839aea · PR result: 25bb297 · Source CI: failure

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: 111.2 KiB
• Claude decoded thread snapshot: 111.9 KiB

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