Improve py::array_t scalar type information

[dean0x7d]

Resolves #632 and the comment in #631 (comment).

Before:

TypeError: incompatible function arguments. The following argument types are supported:
    1. (in: numpy.ndarray[float]) -> object
    2. (in: numpy.ndarray[float]) -> object
    3. (in: numpy.ndarray[complex]) -> object
    4. (in: numpy.ndarray[complex]) -> object

After:

TypeError: incompatible function arguments. The following argument types are supported:
    1. (in: numpy.ndarray[float32]) -> object
    2. (in: numpy.ndarray[float64]) -> object
    3. (in: numpy.ndarray[complex64]) -> object
    4. (in: numpy.ndarray[complex128]) -> object

[dean0x7d]

Updated to also improve the signatures of py::array_t functions. See #631 (comment). Having both float and double reported as float is confusing. Using npy_format_descriptor makes those float32 and float64. This is already the case for Eigen types so I don't think it should be an issue.