RNG in transformed data: interfaces and chains

[Bob Carpenter]

I added _rng functions to transformed data. We still need
to:

* get the RNG reference to the model class constructor
from the interfaces (right now, I rigged it so it'd work
without changing interfaces by creating a fresh RNG in
the constructor if one isn't passed in), and

* figure out how to generate for multiple chains

I was thinking it'd make most sense to use the same transformed
data for each chain. Maybe we generate the random transformed
data first, then advance the seeds based on chain ID? That'd
leave the same results for backward compatiblity.

- Bob

[Sebastian Weber]

Hi!

Hmmm.... I just worked on a case where I needed to have different random numbers per chain. In what you propose would mean that I have to start multiple chains with different seeds which would be ok.

Having the possibility to have the same random numbers for all chains is sensible and then I think that your suggestion to first generate and then split the seeds off for the different chains, makes sense.

Best,
Sebastian