Multicomponent systems incorrectly terminate at peritectic reactions

[bocklund]

Peritectic-type reactions in the ternary go from L+A -> A+B, terminating the current Scheil algorithm. Section 12.8 in Hillert's Phase Equilibria, Phase Diagrams, and Phase Transformations (2nd Edition) may be insightful. Specifically, the idea below might be able to be used to determine when a peritectic region is found and how to update the composition of the liquid so that the liquidus can continue to be followed down the liquidus projection to a eutectic.

[bocklund]

This was actually being caused by a poorly behaved database related to pycalphad/pycalphad#243. There's a test added in #3 that tests this specific case.