Very subtle problem with handling derivatives at zero density

Another bugfix release is needed

Closes #34

Closes #40

Can do binary mixture tracing now!

Seems to be working

Doesn't have the claimed critical temperature; closer before fixing the erroneous coefficient

Ready for Python!

Seems to be fixed now, not sure what the fix was

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

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

Typo in Thorade for dp/drho|T, missing factor of 2

Closing in

Everything seems pretty promising aside from the mutant, which requires additional thought. How can we store the mutant in AllowedModels

Very fast in C++ (less than a microsecond for 4 outputs in the Jacobian), but 10 microseconds in Python for 2. Need to find a way to make it closer to original speed. Perhaps pass in a buffer to be populated.  That should allow the copy to be avoided. It must be the Eigen arrays that are so slow.