fix(pool): fix race in the construction of the sync ConnectionPool lock

fix(pool): fix race in the construction of the sync ConnectionPool lock

The lock used to make `ConnectionPool.open()` thread-safe was itself
constructed lazily in `_ensure_lock()` without any guard. Two threads
racing the first `open()` could each create a different `Lock`, defeat
the mutual exclusion in `_open()`, both pass the `_closed` guard, and
both call `_start_workers()`. The second one tripped
`assert not self._workers`.

Construct the sync lock eagerly in `__init__`. The async lock must stay
lazy because `asyncio.Lock` binds to the running event loop.