Convolution quadrature

I guess it’s possible. But it will involve implementing a lot of smart approximation stuff concerning adaptive refinement of grid, and when to break off the refinement, which I was hoping someone with more expertise and time/resources on their hands had already figured out.

Right now, I think I can do this by just stepping the z value and running something from hcubature on each step. But it would be pretty wasteful, with a lot of recomputation of the same values.