fix: add timeout to Query.close() to prevent CPU hang (#378)

[dicksontsai]

Summary

• Add configurable timeout (5s default) for task group cleanup in Query.close()
• Prevents indefinite CPU hang when tasks don't respond to cancellation

Problem

The Query.close() method could hang indefinitely when tasks in the anyio task group didn't properly respond to cancellation (issue #378). This caused anyio's _deliver_cancellation() to spin at 100% CPU.

Affected scenarios:

• Browser disconnection mid-stream (SSE with FastAPI)
• Subprocess OOM-killed (exit code -9 / SIGKILL)
• JupyterLab extension after calling client.disconnect()

Solution

Set a deadline on the task group's own cancel scope during close():

self._tg.cancel_scope.deadline = (
    anyio.current_time() + self._task_group_close_timeout
)

This approach avoids cancel scope nesting issues that occur when wrapping with anyio.move_on_after() or anyio.fail_after().

Configuration

A new environment variable CLAUDE_CODE_TASK_GROUP_CLOSE_TIMEOUT (milliseconds, default 5000) allows users to adjust the timeout if needed.

Test plan

• Added unit test for close() timeout behavior
• All existing tests pass (118 tests)
• Linting passes

Closes #378