Reimplement selection of correct finite element implementation

So far, we were a bit sloppy about which basis we choose from the UFL counterpart. This commit implements the necessary 1-to-1 mapping.

This fixes #146 (closed)