Add a rule-audit skill: offline static analysis of AI system prompts

Desired Behavior

The extensions catalog should include a skill for static analysis of AI system prompts / agent rule files: detecting contradictory rules, coverage gaps, priority ambiguities, and meta-paradoxes before they cause runtime agent misbehavior. It should run fully offline (no model calls, deterministic), so it can be used in CI.

The rule-audit analyzer (PyPI: rule-audit, MIT) provides this; packaging it as an AgentSkills-format skill under skills/rule-audit makes it usable from OpenHands directly.

Acceptance Criteria

• skills/rule-audit/SKILL.md exists and follows the AgentSkills progressive-disclosure format used by other skills in this repo
• Skill is registered in marketplaces/openhands-extensions.json with a valid category from the README's category list
• Bundled adapter scripts import rule_audit from a plain pip install rule-audit environment without modification
• Marketplace JSON parses as valid JSON

Start by comparing existing skills to the AgentSkills progressive-disclosure format, then inspect the category list in the README and the registration structure in marketplaces/openhands-extensions.json. Add skills/rule-audit/SKILL.md and adapter scripts that import rule_audit from a plain pip install rule-audit environment; done when the marketplace JSON parses and all acceptance criteria are met.