Finally remove anonymous functions in bounds

## Issue
The following construction of bounds is neat but leads to the primary instability of reconstructing EKP priors with JLD2 https://github.com/CliMA/EnsembleKalmanProcesses.jl/blob/4f777393f0e5946ac4ba556fa0ef5949530e0596/src/ParameterDistributions.jl#L176-L185

The reason this arises is a long time ago - there was an un-julia-like decision, which somehow lead to an avoidance of clear dispatch

## A possible solution (Dispatch)

Create the structs
```julia
struct BoundedAbove{FT} <: ConstraintType
upper_bound::FT
end
```
Create the methods
```julia
function c_to_u(c::BoundedAbove, x)
log(c.upper_bound - x)
end

function jacobian(c::BoundedAbove, x)
1.0 / (c.upper_bound - x)
end

function u_to_c(c::BoundedAbove, x)
c.upper_bound - exp(x)
end
```
This would need to be changed in just a few places (exclusively in ParameterDistributions.jl, and FunctionParameterDistributions.jl) where we explicitly call `c.c_to_u` e.g.,
https://github.com/CliMA/EnsembleKalmanProcesses.jl/blob/4f777393f0e5946ac4ba556fa0ef5949530e0596/src/ParameterDistributions.jl#L794-L802

It may also mean that `Constraint` will not be a parameterized type, but rather the abstract type

Start in ParameterDistributions.jl, especially the bounds construction around lines 176-185 and the c.c_to_u calls around lines 794-802, then inspect the corresponding uses in FunctionParameterDistributions.jl. Replace the anonymous-function bounds with dispatch-based constraint structs and methods, while preserving reconstruction of EKP priors with JLD2.

Written by the indexing model from the issue text.