feat(cli): enable concurrent user input during agent execution via asyncio

[phenric26]

This PR addresses the blocking I/O issue in the CLI where users could not provide input while the agent was in the RUNNING state. It refactors the execution loop to support concurrency, allowing real-time feedback and interruptions without needing explicit pause commands.

Key Changes:

Non-blocking Input: Introduced InputManager in openhands_cli/tui.py using prompt_toolkit's async interface.

Concurrent Event Loop: Refactored ConversationRunner.run_concurrent_loop in openhands_cli/runner.py. It now uses asyncio.wait with FIRST_COMPLETED strategy to handle the race condition between user input and agent execution steps.

Safe Execution Wrapper: Added _step_agent_safe to offload synchronous agent steps to an executor, ensuring the main thread remains responsive.

Test Plan (TDD):

This feature was implemented using Test Driven Development (TDD). A comprehensive suite of unit tests was added to cover the concurrency logic:

tests/test_concurrency_flow.py:

    test_input_manager_reads_async: Verifies non-blocking read.

    test_concurrent_loop_exit: Ensures /exit command terminates the loop immediately.

    test_input_interrupts_agent: Validates that user input is processed and sent to the conversation even if the agent is busy.

Related Issue: Fixes #103