(but maybe the compiler elides all the copies anyhow)

Some more testing is needed, but looks promising so far

Big speedup for gradient calculations

Switch critical tracing to use one of the molar concentrations as the fixed parameter when polishing

Expose more critical methods in pybind11 interface

Also not clear if this is the best way to include it.  Probably need the std::variant approach here too.

I think g++ getting confused by copy elision on return of Eigen temporary?