Improve the error messages for unsupported numpy dtypes

[seisman]

Description of proposed changes
Address #2848 (comment).

The current issue is, the error message is helpless and very confusing, so users don't know that np.float16 is not supported:

With this branch, now the error message is more clear:

>>> import numpy as np
>>> import pygmt
>>> x = np.array([1, 2, 3, 4], dtype=np.float16)
>>> pygmt.info(x)
...
File ~/OSS/gmt/pygmt/pygmt/clib/session.py:843, in Session._check_dtype_and_dim(self, array, ndim)
    841             raise ValueError
    842     except ValueError as e:
--> 843         raise GMTInvalidInput(
    844             f"Unsupported numpy data type '{array.dtype.type}'."
    845         ) from e
    846 return self[DTYPES[array.dtype.type]]

GMTInvalidInput: Unsupported numpy data type '<class 'numpy.float16'>'.

The test test_put_vector_invalid_dtype is also updated to test passing arrays with invalid dtypes.

Support for np.float16, np.bool_ as requested in #2848 will be addressed in separate PRs.