feat: Add agent-test-generation skill

[hocokahu]

@zachbai I thought it would be useful to add a skill to generate test cases against all agentic coding. Let me know your thought. Monocle2AI is an open source under the Linux Foundation.

Summary

Adds agent-test-generation — a skill that scaffolds monocle_test_tools pytest tests for Python AI agent apps across LangGraph, Google ADK, OpenAI, Microsoft Agent Framework, CrewAI, LlamaIndex, and Strands.

Covers seven test categories:

1. Agent & Tool Routing — positive + negative tests that the right agent/tool runs for each request
2. Input Validation — verify user inputs are forwarded into agent/tool calls
3. Output Validation — verify outputs contain expected content
4. Performance — token-limit and duration bounds
5. Quality Assessment — dual-mode (see below)
6. Multi-task Orchestration — complex multi-agent requests
7. Individual Agent Testing — each sub-agent in isolation

Dual-mode quality assessment

The quality file works with or without a cloud key:

• Local (default, no key) — deterministic contains_output/does_not_have_any_output assertions plus optional BERTScore semantic similarity (auto-skips if bert_score isn't pip-installed). No network, no LLM call.
• Cloud (OKAHU_API_KEY set) — LLM-as-judge classification via Okahu (sentiment, toxicity, bias, hallucination, etc.). Tests are gated by pytest.mark.skipif so they skip cleanly when the key is absent. The Cloud eval is provided at no cost.