feat: add /ai-observe skill — AI observability and monitoring instrumentation

[HMAKT99]

You ship an LLM feature Monday. It degrades by Friday. Nobody notices.

Traditional monitoring checks uptime and errors. AI features degrade differently: output quality drops, costs creep up, latency increases, the model provider updates the model without telling you. By the time a user complains, it's been broken for days.

What /ai-observe does

You:   /ai-observe

Claude: AI OBSERVABILITY INVENTORY
        Integration    Logging    Metrics    Alerts    Drift Detection
        chat.rb        Partial    None ←     None ←    None ←
        classify.py    Full       Basic      None ←    None ←
        
        COVERAGE: 25% — 3 of 4 integrations are flying blind

        RECOMMENDED METRICS:
        llm_request_latency_ms    Histogram   Alert: p99 > 5000ms
        llm_cost_per_request      Gauge       Alert: > $0.10/request
        llm_output_quality        Gauge       Custom scoring via /eval
        
        DASHBOARD DESIGN:
        Row 1: [Request rate] [Error rate] [Latency p50/p99]
        Row 2: [Daily spend] [Cost per request] [Budget remaining]
        Row 3: [Quality score] [Cache hit rate] [Token efficiency]

Where it fits

gstack instruments the development workflow (plan, review, ship, QA). /ai-observe instruments what you shipped — the AI features running in production.

/review      → code quality before merge
/ship        → push to production  
/ai-observe  → is the AI still working in production?    ← NEW

Only .tmpl committed — bun run gen:skill-docs generates the rest.

Test plan

• .tmpl follows template pipeline — uses {{PREAMBLE}}
• Registered in gen-skill-docs.ts, skill-check.ts, both test files
• bun run gen:skill-docs generates valid SKILL.md
• All existing tests pass with skill added