Add NonManifoldGhostingFunctor and associated SidesToElemMap classes

[jwpeterson]

We often have structural solves involving shells where more than two shell elements meet at a single Edge (see screenshots of some cases where 5 shells meet at a single Edge). In such cases, we don't have complete neighbor information, but it is still possible to correctly ghost the required DOFs when the different neighbors are on different processors. This PR adds a new GhostingFunctor which enables this type of ghosting, as well as several unit tests.

[jwpeterson]

The warning that I'm seeing from the Min clang testing makes no sense

mpicxx -std=gnu++17 -DHAVE_CONFIG_H -I. -I../../tests -I../include  -DNDEBUG  -I/usr/include -pthread -I/opt/petsc/include -I/opt/petsc//include -I/usr/include/tirpc -I../include -I../../contrib/metaphysicl/src/numerics/include -I../../contrib/metaphysicl/src/core/include -I../../contrib/metaphysicl/src/utilities/include -I../../contrib/nanoflann/nanoflann/include -I../../contrib/fparser -I../../contrib/libHilbert/include -I../../contrib/nemesis/v8.11/nemesis -I../../contrib/exodusii/v8.11/exodus/include -I../../contrib/exodusii/v8.11/exodus/sierra -I../../contrib/netcdf/netcdf-c/include -I../contrib/netcdf/netcdf-c/include -I../../contrib/eigen/eigen -I../../contrib/gmv -I../../contrib/qhull/qhull/src -I../../contrib/qhull/qhull/src/libqhullcpp -I../../contrib/triangle -I../../contrib/tetgen -I../../contrib/netgen/ -I../contrib/netgen/build/ -I../contrib/poly2tri/modified   -I../../contrib/gzstream -I../../contrib/sfcurves -I../../contrib/laspack -I../../contrib/boost/include -I../contrib/timpi/src/utilities/include -I../../contrib/timpi/src/utilities/include -I../../contrib/timpi/src/parallel/include -I../../contrib/timpi/src/algorithms/include   -DLIBMESH_IS_UNIT_TESTING  -O2 -felide-constructors -Qunused-arguments -Wunused-parameter -Wunused -ftrapping-math    -fopenmp -Werror -Wall -Wextra -Wcast-align -Wdisabled-optimization -Wformat=2 -Wformat-nonliteral -Wformat-security -Wformat-y2k -Winvalid-pch -Wmissing-field-initializers -Wmissing-include-dirs -Wpacked -Wstack-protector -Wtrigraphs -Wunreachable-code -Wunused-label -Wunused-parameter -Wunused-value -Wvariadic-macros -Wvolatile-register-var -Wwrite-strings  -MT fe/unit_tests_opt-fe_lagrange_test.o -MD -MP -MF fe/.deps/unit_tests_opt-fe_lagrange_test.Tpo -c -o fe/unit_tests_opt-fe_lagrange_test.o `test -f 'fe/fe_lagrange_test.C' || echo '../../tests/'`fe/fe_lagrange_test.C
../../tests/base/nonmanifold_coupling_test.C:69:13: error: loop will run at most once (loop increment never executed) [-Werror,-Wunreachable-code-loop-increment]
            for (auto s : elem->side_index_range())
            ^~~

It's like it doesn't recognize that it's a range-based for-loop for some reason.

We use this construction in many places in the library, though, so no idea why it only -Werrors on this one.

[jwpeterson]

It's the same -Werror on Test clang as well, what the heck?

[moosebuild]

Job Coverage, step Generate coverage on 21554f1 wanted to post the following:

Coverage

	7cda6a	#3975 21554f
	Total	Total	+/-	New
Rate	62.33%	62.34%	+0.01%	78.12%
Hits	72900	72954	+54	50
Misses	44065	44075	+10	14

Diff coverage report

Full coverage report

Warnings

• New new line coverage rate 78.12% is less than the suggested 90.0%

This comment will be updated on new commits.

[roystgnr]

That error does not look like a false positive to me. You've got a for loop that should be iterating from 0 to n_sides. You then have return true at the end of the loop's block. This is equivalent to just setting s=0 instead of the for loop.

[jwpeterson]

@roystgnr I believe I have addressed all your review comments here.