Model Instancing to reduce memory footprint

[Snapex2409]

Idea behind this is the following:

Currently, we save for each micro model its input, output as well as a full copy of the model itself.
IO data cannot easily be optimized, but as this should not be too much usually, certainly less than 1MB per model instance, this should not matter too much.

The critical issue is the full model itself. Depending on its implementation, the internal data structures might use up a significant amount of memory. In the case of certain ROMs this is easily >1MB, leading to very high memory consumption.

However, often the model is stateless. Therefore, the internal data structure do not need to be replicated for each micro-sim, rather can be shared.

Towards this, here is a first implementation, that maintains one "compute" instance for each sharable micro-sim type.

Checklist:

• I made sure that the CI passed before I ask for a review.
• I added a summary of the changes (compared to the last release) in the CHANGELOG.md.
• If necessary, I made changes to the documentation and/or added new content.
• I will remember to squash-and-merge, providing a useful summary of the changes of this PR.