Enable pm.Interpolated to accept symbolic inputs

[kc611]

Original Comment by @ricardoV94 : #4746 (comment)

This issue was faced during refactoring of Interpolated distribution. It uses methods of InterpolatedUnivariateSpline such as integrate() and antiderivative() and these methods aren't directly accessible through the SplineWrapper Op as symbolic operations. (or rather SplineWrapper is not supposed to be used that way as of now.)

[ricardoV94]

@kc611 I edited the title to make it a bit more obvious what the issue / goal is. Thanks for opening the issue!

[jessegrabowski]

I was thinking this might be as easy as some Op wrappers, because we don't want gradients through the interpolation, yes? If we allow symbolic inputs, these should still be root nodes, otherwise we have to deal with stochastic gradients.