fix(knowledge): always write agent trace JSONL for transcript reconstruction (fixes #507)

[mickume]

Summary

Removed the debug-only gate on AgentTraceSink registration so that trace JSONL files are written on every orchestrator run, not just debug runs. Removed the trace_enabled guard in _extract_knowledge_and_findings so transcript reconstruction is always attempted. This makes the trace-based extraction path (113-REQ-1.1) live in production — previously it was dead code because the trace file was never written in normal runs.

Closes #507

Changes

File	Change
agent_fox/engine/run.py	Always register AgentTraceSink in SinkDispatcher (removed if debug: guard); pass trace_enabled=True to factory
agent_fox/engine/session_lifecycle.py	Change trace_enabled default from False to True; remove if self._trace_enabled: guard in _extract_knowledge_and_findings
agent_fox/knowledge/agent_trace.py	Updated docstring ("debug run" → "orchestrator run")

Tests

• All 5249 existing tests pass — no regressions
• Existing tests/unit/knowledge/test_transcript_reconstruction.py covers both the happy path and the fallback when trace is absent

Verification

• All existing tests pass: ✅
• New tests pass: ✅ (existing coverage sufficient)
• Linter / formatter: ✅
• No regressions: ✅

---

Auto-generated by af-fix.