Fully Modularize InfiniteModel Transformation Schemes

[pulsipher]

We currently do allow different optimizer models to be used instead of TranscriptionModels which use a support scheme to solve the InfiniteModel. However, we should fully decouple the attributes associated with transcription from InfiniteModels (e.g., not have it hardcoded to always store parameter supports). Thus, InfiniteModels should be restructured to incorporate only the information needed to represent the infinite-dimensional optimization problem and information needed to reformulate/solve it should be stored in a backend. This will enhance the modularity of InfiniteOpt and greatly enhance the usability of alternative solution approaches.

Such a backend could take inspiration from the way MathOptInterface and JuMP interact. Reformulation/solution schemes like
TranscriptionOpt could ultimately be hosted in their own packages and help break up the size of InfiniteOpt. We also should consider the use case of being able to perform different reformulation schemes over the domains (e.g., transcribe over time and use PCE on the uncertainty).

This will denote a large undertaking that will fundamentally change the structure of InfiniteOpt and will likely parallel the restructuring effort that occurred between v0.1 and v0.2. We can take this on incrementally such that the reformulation schemes become increasingly modular and TranscriptionOpt and all its support-based attributes reside in a fully independent submodule. Once this is achieved, we can move TranscriptionOpt to an external package and finalize our modularized reformulation API.

The fulfillment of this restructuring will make InfiniteOpt a general framework for modeling infinite-dimensional optimization problems and enable the development of diverse solution methodologies.