- Use newer, NumPy-recommended `numpy.random` functions (e.g. replace `rand`
  `random`, `randn` with `standard_normal`, etc.)
- Seed some unseeded/inconsistently seeded tests
- Minor `pytest` usage updates (e.g. use parameterizations, `pytest.raises`, and
  `pytest.warns`)
- Use `np.array_equal` instead of `np.all(... == ...)`

This removes an inconsistency between Numpy and Aesara broadcasting rules, where a variable dimension with unknown shape was always assumed to be non-broadcastable (i.e., different than 1)