Add async timeout and retry logic to agent.act() and agent.think()

[simonlpaige]

Problem

If an LLM provider is slow or hangs, BaseAgent.act() and think() will block indefinitely. There's no timeout, retry, or fallback mechanism.

Proposed Solution

• Add configurable timeout_seconds parameter to BaseAgent (default: 120s)
• Wrap provider calls in asyncio.wait_for()
• Add exponential backoff retry (max 3 attempts) on transient errors (rate limits, timeouts)
• Emit structured log on retry/failure so the pipeline can report which stage failed and why

Bonus

Consider adding a fallback_provider option so agents can degrade gracefully (e.g., try Claude, fall back to local Ollama).