New design for `DataSet` and `Estimator` about feos

[prehner]

Very nice overview!

Two quick comments without having gone through all the code:

• If having a mixture of SINumber and f64 is an issue, we can circumvent that by multiplying, e.g., with MOL/MOL. A bit hacky but could be worth it.
• Personally, I wouldn't mind having an enum with DataSets. We "established" the approach with the equations of state and functionals, and in Python it is already basically treated as enum. But having different values for the associated types would break that concept, wouldn't it?

[g-bauer]

But having different values for the associated types would break that concept, wouldn't it?

Mh not sure if I understand correctly but we could do essentially what we currently do with EosVariant:

• define an enum, say DataSetVariant, that contains all possible DataSet implementations
• implement DataSet for that enumit's not object safe
• possibly use a derive macro to do that?

Or we define the enum without implementing the trait and just write the match statements within the methods of the Estimator that stores a Vec<DataSetVariant>. Both should work.

[g-bauer]

If having a mixture of SINumber and f64 is an issue, we can circumvent that by multiplying, e.g., with MOL/MOL. A bit hacky but could be worth it.

That is part of the problem IMO. The other problem is that the number of return values for a single predict call can be different, i.e. one value for vapor pressure but two values for the chemical potentials when binary VLE's are considered.