feat: Abstract class and constructor for basis function kernels.

[daniel-dodd]

Feature Request

• Define abstract class for basis function kernels.

• Allow users to construct such kernels through specifying combinations of basis functions $-$ how the maths is written.

• The random Fourier feature (RFF) approximation should be subclass of this abstraction.

• If the number of basis functions $m$ is less than the number of datapoints $n$, allow for a mechanism for $O(nm^2)$ time inference (instead of the naive $O(n^3)$ when ignoring the row-rank structure) via the Woodbury identity.