Allow non-Real values in UntypedVarInfo

[mohdibntarek]

It would be nice to allow arbitrary types in UntypedVarInfo instead of just Reals. This should allow Turing to be used with distributions that sample letters, words, trees, images, etc. without having to represent them as an array of reals. The conversion to TypedVarInfo should work fine whatever type is in UntypedVarInfo so this may be as simple as changing Real to Any in a bunch of places.

[cpfiffer]

I would like to see this a lot too -- I could imagine cases where someone had a distribution across bigrams or something and just wanted to work with the actual draws from that distribution.

It seems to me the hangup here is not really so much in DynamicPPL but in MCMCChains and various other spots in the Turing-verse.

[devmotion]

Yep, it would be nice if we wouldn't make any assumptions about the support of the distributions we work with since we just need to be able to sample and evaluate the log pdf.

[trappmartin]

Super exciting. I'm all in for this.

[yebai]

We probably don't need this anymore due to the planned transition to SimpleVarInfo.