More robust handling of multiple solutions for `overpotential`

[rkurchin]

Currently, there can be some wacky behavior when there are multiple valid solutions for overpotential (e.g. before and after inverted region in a Marcus model, or around nonmonotonicities in MHC+DOS models). It would be good to have smarter ways to detect that....in the Marcus case, we know analytically when the inverted region starts, so could try to constrain search to be before it. For MHC+DOS it's a bit trickier, might need to add some stochasticity to the optimizer to find the best minimum and also balance against wanting the lowest-voltage good solution...