ndim = 1 edge case

[CosmoMatt]

We should ensure the code is stable when the number of dimensions is 1. Currently, the JAX functionality is vmapping a function over the samples which is ambiguous in this edge case.

[alicjapolanska]

It turns out this is not just a matter of the vmap, which has been fixed on the 1D_support branch. Additionally substantial revisions to the flow architectures need to be made to support this. For now, I'm adding an issue to add a better error for the 1D case in #313.