Do not implicitly shut down the shared reusable executor

[basnijholt]

What

Follow-up to #497. Runners no longer shut down the internally created executor when they finish; shutdown_executor now defaults to plain False semantics (explicit opt-in), and the docstrings are updated accordingly.

Why

Since #497, every runner with executor=None shares loky's reusable executor singleton. BaseRunner._cleanup used to call executor.shutdown(wait=True) whenever the runner had created the executor itself (shutdown_executor or (executor is None)), which now kills the shared executor out from under any other runner still using it:

loky.process_executor.ShutdownExecutorError: cannot schedule new futures after shutdown

This broke the Read the Docs build of #497 while executing tutorial.advanced-topics.md (a cancelled goal-less runner's cleanup raced with the next runner's await runner.task). The race predates #497 on macOS/Windows, where loky was already the default, but docs only execute on Linux so it was never hit.

Idle loky workers shut down on their own after their timeout, and loky cleans up at interpreter exit, so nothing leaks.

Testing

• Minimal repro (cancel a goal-less runner, then run a second runner) raised ShutdownExecutorError before, passes now.
• Full suite passes on Python 3.14.5 (except the pre-existing test_to_dataframe failures that also occur on main).
• pre-commit passes.