Fix discretization of flux at electrode/separator interface

[Edwin Khoo]

The current implementation of the flux discretization at an electrode/electrolyte interface is presumably 1st-order while the discretization elsewhere is 2nd-order, so there is a possibility the accuracy of the solution everywhere degrades to 1st-order. We should discretize the flux taking into account non-uniform grid spacings on both sides of the interface.

[Raymond Smith]

This may be a nice addition, but it could also lead to common issues with wide stencils like instability or decreased robustness. Maybe not, so it's probably worth trying. However, I also don't think it's that important. The other issue is that we've never confirmed we're getting second order convergence in the case of equal discretization size for electrodes and separator. I think we should do that first, and it could be limited by the way we integrate over the reaction term (maybe?). If we get second order convergence with uniform spacing, we could try this.

[Dan Cogswell]

Starting in bc7f166, values at cell faces are now weighted by the size of the neighboring cells.