Allow scale to compose with transformations

[WardBrian]

This is a much more limited alternate take on #947. Basically, separate out offset and multiplier from the idea of a transformation, and allow them to be composed, but not arbitrary other compositions. This aligns more directly with #659

Submission Checklist

• Run unit tests
• Documentation
  • If a user-facing facing change was made, the documentation PR is here: TBD

Release notes

Offset/Multiplier scaling can now be applied to constrained data types, like real<lower=0><offset=1>. When constraining, the scaling is done first so that the result is truly of the constrained type.

Copyright and Licensing

By submitting this pull request, the copyright holder is agreeing to
license the submitted work under the BSD 3-clause license (https://opensource.org/licenses/BSD-3-Clause)

[WardBrian]

@SteveBronder - This is the proof of concept of the alternative version. If you ignore the multiarch build problem ATM it compiles all the models fine, including the following:

data {
   int<lower=0> N;
}

parameters {
    array[3] real<lower=0><offset=10> loweroffset;
    real<lower=0, upper=1><offset=1, multiplier=10> allfour;
    simplex<offset=1>[10] offset_simplex;
    positive_ordered<multiplier=10>[N] multiplied_ordered; 
}

We can throw this into cmdstan and start doing some actual modeling tests next week

[WardBrian]

The implementation here should be good to go, @bob-carpenter and I need to work on some example models and the doc for it.

[SteveBronder]

Is this waiting on review?

[WardBrian]

There was a lot of discussion in #659 about whether this was the right thing to do or not, which is the bigger blocker I believe

[WardBrian]

This has fallen too far behind, so I'm going to close it for now