Change function space hashing API to accept an L^p-space similarity instead of corresponding l^p-space similarity

[kernelmethod]

Currently, we construct a new function-space hash function along the lines of

hashfn = MonteCarloHash(ℓ2, μ, args...; kws...)

This isn't very intuitive; it would make more sense to use L2 instead of ℓ2 since the former is a distance on a function space and the latter works over discrete vectors. ChebHash and MonteCarloHash should be modified so that they accept the function-space similarity that they are using rather than the corresponding discrete-space similarity.