`Bijectors.jl` for random variable transformations?

Transforming from "constrained" (physical space) distributions to "unconstrained" (EKI-space, `Normal`) distributions is a core component of our interface:

https://github.com/CliMA/ParameterEstimocean.jl/blob/cc601ec7cf1c413b5d6d058f661c0a9105301fc1/src/Parameters.jl#L216

@briochemc points out that it might be possible to use

https://github.com/TuringLang/Bijectors.jl

which could potentially allow us to support any prior `Distribution` without having to code the transformation ourselves? (I'm not sure this actually makes sense... but leaving this here in case we want to dive deeper in the future).

Start by reading the transformation code in src/Parameters.jl around line 216, then review the proposed TuringLang/Bijectors.jl integration. Determine whether Bijectors.jl can support the constrained-to-unconstrained distribution transformations needed here and whether it would enable arbitrary prior Distribution types. Done means documenting a clear integration decision and scope.

Written by the indexing model from the issue text.