fix(prime): wait for async child continuations

[rynfar]

Problem

Prime daemon sessions completed Pylon's canonical turn at the first native RunCompleted event. A depth-1 child could finish afterward and resume its parent internally, but Pylon had already emitted a false missing-final warning, completed the turn, dropped the resumed final answer, and left the client in a stale working state.

Fix

• use Prime Agent 0.8.0's public waitForHeadlessCompletion({ waitForRlmQuiescence: true }) boundary after each Pylon-owned daemon prompt
• correlate each FIFO quiescence marker with the canonical turn/input token and daemon connection generation
• re-arm and serialize barriers after steering or follow-up input, ignoring stale markers from earlier inputs or turns
• fail the turn and dispose the native session when a barrier rejects or reconnect makes event continuity uncertain
• compute terminal usage from validated cumulative stats so child and continued-parent usage is included
• retain legacy response-boundary behavior when the authoritative barrier is unavailable
• document the completion, reconnect, usage, and version-fallback contracts

Closes #80.

Verification

• combined Prime/MCP regression suite across eight focused files — 274 passed
• vp run -F t3 typecheck — passed
• targeted lint on all seven changed TypeScript files — 0 warnings, 0 errors
• formatting and git diff --check — passed
• real isolated Pylon web client on 7e3529c3a with Prime Agent 0.8.0 — one depth-1 child completed after the root response boundary and resumed its parent with PYLON-ASYNC-FINAL-7E35; SQLite recorded one completed canonical turn, both root response segments, no runtime error or false missing-final warning, and the terminal reported cost included the child-inclusive cumulative usage delta
• fresh PR CI on 7e3529c3a — 11 checks passed, 3 expected skips, 0 failures

Implemented by GPT-5.6 Sol using Prime Agent.

---

^{Need help on this PR? Tag @codesmith 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	13.3 KiB	13.3 KiB	−2 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.4 KiB	−4 B (−0.1%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.6 KiB	55.6 KiB	−44 B (−0.1%)	66.4 KiB	✅
Codex	Live turn messages	11	10	−1 (−9.1%)	21	✅
Claude	Total thread wire	13.3 KiB	13.3 KiB	+37 B (+0.3%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	−13 B (−0.2%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+50 B (+0.8%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.3 KiB	56.4 KiB	+44 B (+0.1%)	66.4 KiB	✅
Claude	Live turn messages	9	10	+1 (+11.1%)	21	✅

Baseline: 14bad65 · PR result: 7e3529c · 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: 109.4 KiB
• Claude decoded thread snapshot: 110.1 KiB

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