perf(server): reduce OpenCode streaming memory

[rynfar]

Long OpenCode sessions retained tool output that Pylon never reads and repeatedly scanned text parts from earlier messages. Streaming progress also serialized growing tool output into native event logs.

Port four upstream performance fixes (#9684, #9689, #9738, and #10116): drop unused tool history and tool-part caches, index text parts by message, and omit repeated running-tool progress from native logs. Final tool results, errors, client events, and native rollback remain intact. The port preserves Pylon’s session incarnation ownership and turn admission. It does not introduce upstream’s separate token accounting feature.

Validation: 194 focused OpenCode adapter, event logger, and provider service tests passed; server package typecheck and targeted lint passed. Regression coverage includes late message metadata, completed-text edits, reconnects, part removal, and long histories.

Model: GPT-6 Astra. Harness: Codex.

---

^{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	−15 B (−0.1%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	−9 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.5 KiB	−6 B (−0.1%)	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	13.8 KiB	13.6 KiB	−146 B (−1.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+2 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.5 KiB	−148 B (−2.2%)	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: c966815 · PR result: 33378cc · 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: 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.