Consider switching to linear mean for CHR flux

[Raymond Smith]

At least in the case of the constant voltage graphite simulation from the Guo et al. paper, the simulations are much faster using linear means for the concentration flux prefactor. It seems overall less reasonable to use linear means, but also less problematic to do so only in the solid where there are no electric fields driving the flux.

[Dan Cogswell]

Starting in bc7f166, transport properties are interpolated to faces using weighted harmonic mean, and concentration using a weighted linear mean.