Clean up Hilbert v SingularIntegral

[dlfivefifty]

SingularIntegral(S,0) is really different than SingularIntegral(S,1): it's real valued, is not (x-z)^0, etc.

I think LogKernel and SingularIntegral should be different types.

[dlfivefifty]

I changed my mind on this but we should flip the sign of Hilbert: this would make it consistent with https://en.wikipedia.org/wiki/Hilbert_transform and the derivative of the log kernel. So

SingularIntegral(k) === Derivative()^k * SingularIntegral(0)