Switch TP VLE solver to Powell's dogleg method. Might be just a little bit more reliable, but hard to tell.

Add routine for TP flash for binary mixture, to solve for molar concentrations (and therefore mole fractions)

Not as fast as it could be, suspect the string conversion is not helping

Compilation of user code is very fast now

Could also be helped by perfect branch prediction

The use of void* in the interface is not my favorite solution...

See https://github.com/usnistgov/teqp/issues/9

More testing is still needed to identify other issues

Bump version

Stopped iteration if dx values are all < 0