Closes #51.
Adds the option to assemble local matrices via computeAndStoreLocalMatrices(). Adds the option to load a local matrix instead of assembling it on-the-fly during Apply(). With adaptive Galerkin coarsening in mind, this is done in a selective manner: a map of PrimitiveIDs can be passed to the operator; only for the specified primitives the local matrices will be assembled and applied from memory.