Use `scope` effect handler for distributions with multiple RandomVariable arguments

[damonbayer]

https://num.pyro.ai/en/stable/handlers.html#scope

For example, a Negative-Binomial distribution, could be given one name, "Infections" and its parameters automatically renamed "Infections_mean" and "Infections_dispersion" or similar.

[damonbayer]

I don't think this is worth implementing for any of our current models, but could be useful at some point in the future.

[dylanhmorris]

@cdc-mitzimorris

[dylanhmorris]

I have been thinking about this while implementing #696 and I think one approach could be to define a scope() method for the base RandomVariable class that always uses the RandomVariable name as the prefix and uses a standard divider (I vote _). Then any RV that wants to call other RVs within its sample method can do

with self.scope():