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

Tiny updates, nothing too serious

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

Heading towards pure VLE derivatives

The build matrix is already crazy

Closes https://github.com/usnistgov/teqp/issues/30