Finite Element Formulation follow up.

The finite element formulations PR #1014 needs to be extended a bit. The following tasks should be completed soon:

• Provide shape function gradient methods for the triangle and quadrilateral. While they are not currently needed, this would complete the functionality.
• Recast the FiniteElementDispatch to use a generic dispatch function as suggested by @klevzoff. Propagate this generic dispatch function throughout the code. (constitutive dispatch, etc). The original comment:

I would like to reiterate my comment https://github.com/GEOSX/GEOSX/pull/1014#discussion_r463254636
With a variadic recursive dispatcher class you could make this whole function be just

GenericDispatch< H1_Hexahedron_Lagrange1_GaussLegendre2,
                 H1_Wedge_Lagrange1_Gauss6,
                 H1_Tetrahedron_Lagrange1_Gauss1,
                 H1_Pyramid_Lagrange1_Gauss5 >::dispatch( input,
                                                          std::forward< LAMBDA >( lambda ),
                                                          "finiteElement::dispatch3D()" );

and not worry about naming of temporaries, etc.

Originally posted by @klevzoff in https://github.com/GEOSX/GEOSX/pull/1014

• Determine whether or not we want to use the ObjectCatalog in place of the FiniteElementDiscretization::Factory.

Start by reviewing PR #1014 and the referenced discussion about generic dispatch. Trace FiniteElementDispatch, constitutive dispatch, the triangle and quadrilateral shape-function APIs, and FiniteElementDiscretization::Factory versus ObjectCatalog. Done means completing the listed gradient methods, propagating the dispatch design, and deciding the catalog approach.