Implement narrow distribution family cases as their own distributions where more optimal #335
Description
#332 calls out how Bernoulli(p) is a wrapper for Binomial(p|n=1) which led to some inaccuracy in cdf evaluation.
We should implement the special case where possible.
For Bernoulli, remaining features include mass function, survival function, descriptive statistics for the realization, its own error type.
Erlang may not benefit from this as much.
_Some of this motivates exposing the functions that don't require persistent data and having the methods wrap those. gamma::pdf would be useful in <Erlang as Continuous>::pdf, however it wouldn't be as apparently reusable for things like 'Empirical' that have associated structured data.