Refactor tests to use floating points for data (not `np.ones` or `np.zeros`)

[tomvothecoder]

There are some tests (e.g., TestDepartures) that use np.ones or np.zeros, so the output of these operations won't produce outputs the discern correctness of behavior.

Instead, we should have tests run with between 3-5 input floating point values when performing operations such as averaging so that we can reliably test APIs.

https://github.com/xCDAT/xcdat/blob/b2be983598e04a63e6ca4674e10ccdc006251f72/tests/test_temporal.py#L1414-L1428