Retire the previous background worker pool when resetting the executor #404

Resetting Cachier's background executor replaces its reference without calling shutdown() on the previous ThreadPoolExecutor. An application that updates the worker setting can leave old idle workers alive and old pools still accepting submissions until garbage collection cleans them up.

Reproduced on master e5fd990, Python 3.12.14, memory backend / core worker utility:

from cachier.core import _get_executor
old = _get_executor()
old.submit(str, 'start a worker').result()
_get_executor(reset=True)
old.submit(str, 'still accepts work').result()  # unexpectedly succeeds

The previous executor should stop accepting new submissions while its running and queued calculations finish. Reset should not wait for long-running cache calculations. A regression test reproduces the missing retirement; a proposed fix also preserves the original pool if constructing its replacement fails. This isolates item 15 in #367.