Rework scaling transformation to avoid renaming components

[jsiirola]

Summary

Renaming components in the scaling transformation can be problematic (see, e.g., #2678). We should rewrite the transformation to avoid renaming components.

Rationale

#2678 pointed out several limitations with the scaling transformation. In particular:

• renaming components does not play well with References, especially when only scaling / transforming a part of the model. We should consider reworking the scaling transformation to be out-of-place, deactivating the original Constraints and creating new Variables / Constraints in a transformation block.
• out-of-place transformations (e.g., _create_using() could be made more efficient to leverage _apply_to and just return the new transformation block as the model
• We should explore generalizing the AutoLinkedBooleanVar concept from GDP to generate a generic "update callback" mechanism for components like Vars
  • (note that we could also do something similar for Sets, which is a "feature" that has been frequently assumed to exist).

Description

Additional information

[Robbybp]

Note that as of #1596, _apply_to, which is called by _create_using, has a rename=True optional argument, which you can set to False to not rename, allowing the scaling transformation to work (without surprises) on models with references.

At the time, the "fix" implemented by #2678 was not possible as ReferenceDict did not exist. However, I remember discussing that references, as essentially symbolic links, should be broken by renaming, and that we deliberately avoided trying to hack our way around this (the way to do this at the time would have been to modify the internal slice data structure - creating an "equivalent" ReferenceDict is less offensive, although still breaks component is renamed_component comparisons and does not preserve indexing sets).