STLconductor caching fails with mesh refinement solver

[cchall]

Because STLconductor caches intercept data by mglevel STLconductor does not work with the MRBlock solvers. Child meshes use the same STLconductor object as parents so there are repeated calls to the same mglevel, but with different mesh shapes. Which is fortunate since each new mesh should have a new intercept calculation.

Assuming that Warp is caching its intercept data (which seems to be the case) then the easiest solution is to change the caching key in STLconductor to be the intercept's id.