feat: add mock LLM mode to run agent without API key

Add a mock mode to LLMClient (BASEAGENT_MOCK_LLM env or mock= param) that
bypasses the API-key requirement and returns a deterministic response with no
function calls, so the agent loop completes after one self-verification turn
while still executing real environment commands. This enables end-to-end
pipeline verification without any LLM provider credentials.

- src/llm/client.py: _truthy helper, mock branch in __init__, _mock_chat(),
  chat() short-circuit, None-safe close()
- agent.py: mock-aware run() honoring BASEAGENT_MOCK_LLM from env/context
- tests: mock-run-without-key test; updated kwargs assertion (mock: False)