LHS for dynamic (no fault) appears to be double counted along process boundaries

[baagaard-usgs]

Describe the bug

Lumped LHS that is inverted for dynamic (no fault) appears to be double counted along process boundaries and across cells.

Desired solution

• Use quadrature that results in a diagonal mass matrix for the integration.
• Replace use of the Jacobian action and row sum with conventional integration.

Questions

• Can we use a vector instead of a matrix to avoid unnecessary memory allocation. We don't want to allocate enough memory for a sparse matrix.

[baagaard-usgs]

PETSc has a function for setting the quadrature type. The options are conic, minsym (default), and diagsym. We will use diagsym for a lumped mass matrix.