Add synthetic coordination agent trace

[CalibratedGhosts]

Summary

Adds a small synthetic coordination_agent trace for the agent-workflow trace collection discussed in LMCache/LMCache#1826.

The trace models a coordination agent that repeatedly:

• orients on a public task,
• builds a source map from public artifacts,
• applies policy/privacy checks,
• writes a concise receipt,
• hands off the next owner-visible follow-up.

It follows the pyCodeAGI style rather than raw log dumps: README.md, trace.jsonl, and a deterministic dependency-free generate_trace.py.

Privacy / Scope

This is intentionally synthetic and representative. It does not read or include private transcripts, private messages, raw logs, credentials, personal data, hidden reasoning, or environment state. The README states this explicitly.

I also avoided adding a performance chart or LMCache speed claim, since I did not run the full cache analysis locally.

Validation

• python3 coordination_agent/generate_trace.py
• JSONL line validation: 15 rows, 3 sessions, required keys present
• python3 -m py_compile coordination_agent/generate_trace.py
• git diff --check
• local secretish scan over coordination_agent/

Analysis caveat: local environment is missing matplotlib, so I did not run prefix_analysis.py.