SMT repository refactorization for general-purpose experiments

[antoniorv6]

Hey there, I am opening this discussion to know if this should be approached or keep things as-is.

This is a more general-purpose issue. Now, the repository features datasets that are exclusively tied to the ones that have been tested in our papers (which makes sense for the sake of replication). However, if someone needs to use their own dataset, they need to copy/paste their own code into data.py and avoiding lots of functions that they may not need if Humdrum **kern is not considered for experiments.

My question here is: Should we make the SMT project ready for more generic experimentations? That is, making a datasets factory where users may inject their own code files and work exclusively with their code aided by generic integration functions or should we keep the policy where users should fork their own versions of the SMT and handle the code as they may want?

I also have the same wondering with the SMT model itself, although I think it is better that users make their own versions through huggingface-based fodler organization.