refactor: Make agent mode interruptions more robust

[szmania]

Summary

This PR makes agent mode interruptions more robust by ensuring that long-running operations (linting, command execution, streaming responses, and LLM completion calls) can be properly interrupted via the existing interrupt_event mechanism.

Changes

Core Interruption Infrastructure

• cecli/helpers/coroutines.py: Added interruptible_async_generator() - a utility that wraps async generators to make them interruptible via an asyncio.Event, enabling cancellation of streaming LLM responses mid-generation.

Linting (cecli/linter.py)

• Made Linter.lint(), Linter.run_cmd(), Linter.py_lint(), and Linter.flake8_lint() all async so they can be interrupted.
• Linter now accepts an interrupt_event parameter and uses run_cmd_async instead of blocking subprocess.run() calls.
• When interrupted during linting, returns None (graceful handling) instead of crashing.

Command Execution (cecli/run_cmd.py)

• Added run_cmd_async() - an async version of run_cmd that uses asyncio.create_subprocess_shell and monitors an interrupt_event to terminate processes early when interrupted.

Agent Coder (cecli/coders/agent_coder.py)

• Linting in the agent loop now uses coroutines.interruptible() to allow interruption.
• Waiting for background commands now uses coroutines.interruptible() instead of a blocking asyncio.sleep().

Base Coder (cecli/coders/base_coder.py)

• Streaming responses: The show_send_output_stream() method now wraps the async generator with interruptible_async_generator(), allowing Ctrl+C to interrupt streaming LLM responses mid-generation.
• Message formatting: Uses coroutines.interruptible() to allow interruption during format_messages().
• Lint editing: lint_edited() is now async and handles asyncio.CancelledError gracefully.
• Run commands: Uses run_cmd_async() instead of asyncio.to_thread(run_cmd, ...) for interruptible command execution.

Run Command (cecli/commands/run.py)

• Switched from asyncio.to_thread(run_cmd, ...) to run_cmd_async() for interruptible execution.

Models (cecli/models.py)

• LLM completion calls (litellm.acompletion) now use coroutines.interruptible() to allow interruption during API calls.

Testing

• All existing tests pass.
• Manual testing confirms Ctrl+C properly interrupts linting, command execution, and streaming responses in agent mode.