Spectral distribution and domain of kernel input

I'd like to have a function spectral_distribution which takes a kernel defined over R^d and returns in spectral distribution. For example

function spectral_distribution(k::SqExponentialKernel)
  MvNormal(dim(k), ones(dim(k)))
end

However, there is one big API incompatibility with such a method: currently, there is no way to determine what domain a KernelFunctions.jl kernel is defined over. In particular, there's no way to tell whether a given kernel is defined over R^1 or R^3 or R^100. The underlying issue is that Distances.jl does not distinguish this, and KernelFunctions.jl largely works on top of that API. This is also an issue for #9 and #10.

I suggest instead adding the domain as an explicit parameter to the kernel.

Start by reviewing the KernelFunctions.jl kernel API and the Distances.jl assumptions that prevent a kernel's domain dimension from being identified. Compare the proposed spectral_distribution function and explicit domain parameter with the concerns in issues #9 and #10. Done means the API direction is resolved for determining kernel domains and supporting spectral distributions.