Panzer: Commit c468d41b Breaks Charon

Created by: jmgate

@trilinos/panzer

Expectations

After @bathmatt ripped out the std::vector<int> of LIDs in c468d41b, such that we'll only use the Kokkos::View version instead, everything should behave the same.

Current Behavior

Charon has a particular example involving Joule heating and a heterojunction (I don't pretend to understand the example, but it seems the combination of those two factors is significant), where Panzer_ScatterDirichletResidual_Epetra_impl.hpp runs into a problem. In the Jacobian specialization of evaluateFields(), this bit throws because err is 2:

if(!preserveDiagonal_) {
  int err = Jac->ReplaceMyValues(row, cLIDs.size(), scatterField.dx(), &cLIDs[0]);
  TEUCHOS_ASSERT(err==0);
}

I put the old vector of LIDs back in place to compare, and here's what I'm seeing for the last cell in the workset before everything dies:

rLIDs = {83895, 83896, 83897, 83898, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
rLIDsVec = {83895, 83896, 83897, 83898}
cLIDs = {83895, 83896, 83897, 83898, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
cLIDsVec = {83895, 83896, 83897, 83898}

It seems the Kokkos::Views aren't the right size, but for all prior cells in the workset, the vector and View versions all have 16 entries. @bathmatt, @eric-c-cyr, any idea what's going on here or how to fix it?