✨ feat: improve v0.3.1 agent investigation

[itdevwu]

Summary

• turn the canonical Skill into a survey-to-hypothesis performance investigation playbook
• bundle deterministic schema-v2 JSONL analysis for kernels, copies, busy union, overlap, streams, gaps, and launch variants
• add version, bounds, multi-process, JIT, perturbation, and profiler-handoff guidance with exact-policy examples
• codify evidence verification and release discipline in AGENTS.md
• add a scoped private-reporting security policy and explicit Linux trust boundary
• prepare the version-matched v0.3.1 package and installation documentation

Verification

• just fmt-check
• just lint
• just test (outside the sandbox for ptrace)
• just test-agent-contract
• just test-install
• just test-skill-install
• Skill quick_validate.py

Closes #19
Closes #20