Fix destroy for asyncio (#451)

This PR fixes a bug where `DBOS.destroy()` shuts down the executor pool
but leaves it attached to the event loop. As a result, if DBOS is
re-initialized and async functions are invoked directly, the loop still
tries to use the destroyed pool and raises a runtime error.

The fix is to let the wrapper functions optionally accept a `dbos`
instance. This makes sure the event loop is always configured correctly,
even when async workflows are called directly after re-initialization.

Added a test for destroy semantics for asyncio.