Ready for Python!

Using sigma in A seems to be necessary to avoid underflow in some numerical calculations, as is done in PC-SAFT

Still need the model and so on, but working

n[21] was -0.99218575e+1 instead of -0.99218575e-1

clang on mac doesn't mind

Also test dipolar with JSON

The references seem to be getting broken. There must be an implicit copy somehow.

Added into the coeffs structure, nice interface

This gives access to multiple output parameters, for instance eta, needed for polar contributions

Need to decide on how to handle the model composition with the chain models

Refactoring of the SAFT contributions is going to be needed

And some tests

Can also specify them as JSON-formatted string, to make fitting a bit easier, for instance

And get the parameters from the class

* Try to break up the C++ interface into smaller chunks

In order to make compilation less RAM intensive

* Inline the makers, also add the other sources

* Tidying up the splitting

Problem was mcx missing sqrt

Recommended by static analysis in Xcode

(no implicit conversion of nlohmann::json to filesystem)