Random features for Gaussian kernel approximation functions provide no sigma parameter explicit choice, nor associated paramsel method

[raffaello-camoriano]

When working with random Fourier features (RFs) for the approximation of a Gaussian kernel, it would be very useful to have the following functionalities:

• A field of opt to set the sigma parameter used in the random feature mapping.
• A paramsel method such as paramsel_siglamho, but working for RFs.