Implement a pending status for tasks.

[tobiasraabe]

Partially closes #98.

Inspired by https://stackoverflow.com/questions/56587166/incorrect-number-of-running-calls-with-processpoolexecutor-in-python/56587872#56587872.

[tobiasraabe]

The PR includes an implementation where the status of tasks is shared with some shared object. No clue why the pickle error exist when running the test suite with tox, but not without it.

[tobiasraabe]

Merged main into this branch, then switched pending→running tracking to a start-event queue:

• manager queue for processes/loky
• SimpleQueue for threads
• wrappers emit start events; main loop drains queue to mark RUNNING
  I also restored task submission to "n_workers - len(running_tasks)" to preserve max_failures and priority behavior.

Tests:

• uv run --group test pytest tests\test_execute.py -k "stop_execution_when_max_failures_is_reached or task_priorities"
• uv run --group test pytest