[New feature] Faster sampling for small-dimensional cops

[lrnv]

Small dimensional archimedean copulas (in particular d=2) are really usefull and sometimes have specific samplign schemes taht are much faster than what we are doing right now.

We could try to implement some of them to fasten the sampling routines.

[lrnv]

Maybe splitting the frailty/non-fraitly geenrators like this https://discourse.julialang.org/t/one-type-with-branches-or-a-union-of-types/132380 could be usefull