Fix the bug of converting Python sequence of datetime-like objects

[seisman]

Cherry-picked from #3758. Tests will be added in #3758.

Fixes #3757.

Decided that #3758 (comment) might be worth adding

[weiji14]

Originally posted by @seisman in #3758 (comment)

Maybe we should extend the unit test to include pandas.Timestamp? E.g. adding these lines:

    # list of pandas.Timestamp
    x = [pd.Timestamp("2020-1-1"), pd.Timestamp("2021-1-1")]
    y = [10.5, 11.5]
    fig.plot(x=x, y=y, style="h0.2c", pen="1p")

See the explanations above. As long as _to_numpy can convert a datetime-like object into np.datetime64 array correctly, PyGMT should be able to plot/process it. So, I think the newly added tests are enough.

Still think we should have an integration test of some sort for Sequence[pd.Timestamp], or maybe Sequence[pyarrow.TimestampScalar]. We should cover both np.object_ and np.str_ type arrays, in case the internal virtualfile_from_vectors->vectors_to_arrays->_to_numpy logic ever changes.

[seisman]

OK, added a test in 892f444.