WIP: Chemical potential conditions support

[richardotis]

This PR adds a new, flexible constraint system via JIT-compiled functions. This new flexibility is used to add support for chemical potentials as conditions in an equilibrium calculation.

[richardotis]

Thank you for the review. The quality of the code has been significantly improved.

1. Yes, we can do grids of chemical potential calculations now. The approach for phase diagrams is largely up to you since I see no reason to put effort into the original approach with fast mapping around the corner. One question I don't know yet is the behavior for chemical potential conditions in the very dilute case (~1e-20 and below). I have not tested that case.
2. Yes, I agree. I would like to defer this until the next advanced conditions PR so I can confirm no more design changes will be necessary in the short term.

In addition to another round of review in light of these revisions, I'd also like your concurrence explicitly on the proposed temporary SkipTest for test_dataset_can_hold_maximum_phases_allowed_by_gibbs_phase_rule.

While making changes to the solver to improve convergence, I found that the results of this test became unreliable and platform-dependent, requiring minute changes in the calculation temperature to resolve. I now believe that this test working at all was a bug, and for the given set of conditions equilibrium should never return more than two phases. Instead, the test should be rewritten to use NP conditions with temperature as a free variable. I propose that we skip this test for now and rewrite it for the NP PR.

[bocklund]

In addition to another round of review in light of these revisions, I'd also like your concurrence explicitly on the proposed temporary SkipTest for test_dataset_can_hold_maximum_phases_allowed_by_gibbs_phase_rule.

While making changes to the solver to improve convergence, I found that the results of this test became unreliable and platform-dependent, requiring minute changes in the calculation temperature to resolve. I now believe that this test working at all was a bug, and for the given set of conditions equilibrium should never return more than two phases. Instead, the test should be rewritten to use NP conditions with temperature as a free variable. I propose that we skip this test for now and rewrite it for the NP PR.

I think we have discussed this offline in the past and that freeing up another condition (T in this case) is the correct approach.