feat: add advisory action risk check skill

[gethsun1]

Summary

Adds an advisory action-risk-check skill to OmegaClaw.

The skill deterministically evaluates an intended tool action and returns:

• decision: allow, review or block
• risk: low, medium, high or critical
• a stable reason code

Implementation

• Registers the skill in src/skills.metta
• Adds a MeTTa-to-Python bridge
• Adds deterministic classification in src/action_risk.py
• Registers the command with the response parser
• Prevents sensitive targets from being echoed
• Adds documentation and table-driven tests

Safety model

Classification precedence:

1. Destructive
2. Sensitive
3. State-changing
4. Read-only
5. Unknown

Unknown actions default to review/medium.

Validation

• Python unit tests: passed
• Python compilation: passed
• Parser checks: passed
• git diff --check: passed
• PeTTa integration test: pending containerized verification

Limitation

This version is advisory and must be called explicitly. It does not yet
intercept every OmegaClaw action before dispatch.

A future phase could introduce a pre-dispatch enforcement hook.