Specify scalar type as numpy type or numpy type name

[miklos1]

It would be better to receive scalar type in the parameters dict not as a C99 typename string, but rather as a numpy type (np.float64) or numpy type name ('float64') -- they are easily convertible to one another, as well as to dtype instances. The latter helps deciding complex mode easier (dtype(name).kind == 'c') as well as looking up default precision.

This representation of scalar type could easily be mapped to a C type name through a predefined dict when it comes to code generation. Alternatively, one could map to C++ type names when needed, e.g.,
double complex is C99 and does not work in C++, which is problem when the generated code needs to be compiled as C++ because of Eigen, for example.

The numpy type representation is also required when interfacing loopy.