Improve Agent Failure Handling in OrchestratorAgent

[YanCotta]

Description:

The _handle_agent_failure method in src/agents/async_orchestrator.py has a very basic retry logic for the search agent. This should be expanded to a more robust and generic error handling and recovery strategy that can be applied to all agents.

Acceptance Criteria:

[ ] Implement a configurable retry mechanism with exponential backoff for failed tasks.

[ ] The orchestrator should be able to handle failures from any agent in the workflow.

[ ] If a task fails after multiple retries, the orchestrator should gracefully terminate the workflow and report the failure.

[ ] The new error handling strategy should be documented in the ARCHITECTURE.md.

Files to check:

src/agents/async_orchestrator.py