Speeding up face integration loops

[stefanhenneking]

Currently, many routines (such as elem for DPG formulations) use inefficient face integration by iterating over more shape functions than necessary. Shape functions should only be evaluated on the particular face being integrated over. This can be accomplished by modifying the norder array in a way specific to the face being integrated before calling the shape functions package. This is currently implemented in the hpinterp face routines that integrate over faces.