As part of this work the headers SetupRegionMatrix and SetupRegionHierarchy are partially rewritten to fix the Xpetra path.
Additionally a coarse grid solve using Tpetra/Amesos2 is added and gets resonable convergence improvements.
Since we have things working, we should protect them with tests.
This provides a small first test in 2D that performs correctly with coarse grid solve.

1) issue warning for missing coarse solver
Coarse level solver has not been migrated to Xpetra yet.
For now, we just skip it and don't do anything on the coarse level.
However, proc 0 writes a warning message to the screen.

2) Fix map when reading input matrix
Use the composite map instead of creating a linear map when reading
the input matrix.
Remove obsolete function argument

3) Specify example via command line:
The HHG driver reads various input files. The exact example has to be
known in order to setup the structured region MG. Now, the exaxt
name of the example has to be passed via a command line arguement
"--caseName=..."

Add include guards
Move utilities to utils file

Part of #4084.

This is a starting point only, actually some difficulties with the LIDRegion, appData and myRegions arrise quickly preventing further progress.
Also removing unsued variables from research region_algorithms/Driver.cpp

After creating the hierarchy, we need to now its number of levels in
order to run the HHG-specific V-cycle. For now, we extract this
information from the HHG prolongator data structure.

Part of #4084.

Note: also fix Matlab syntax. Use ' instead of " and move LID widget to Utils file

The idea is to seperate some capabilites from the actual driver in order to develop more generic HHG headers.
These headers can then be included in various driver instead of the current in-place use.
This new header defines hierarchy setup functions and the vCycle.
It could be renamed RegionMG but let's think about that later.
It exposes a better split among various functions that are used in the Region MG algorithm.
A few const qualifiers have been modified to be more accurate.

- Direct solver on coarse level is missing.

Closes #4084, #4193.

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Xpetra: fix header install problem
PR Author: jhux2

Fixes #5011.

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Tpetra: fix clang compilation of test using clone
PR Author: kddevin

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Giving group read and execute perms for 'decomp' in binary dir (ATDV-156)
PR Author: bartlettroscoe

Xpetra: start of ETI system

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Intrepid2: simple workaround for nvcc compiler bug (issue #4963)
PR Author: mperego

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Trilinos:  Deprecate four packages (second attempt)
PR Author: kddevin

The strategy that I am using to install Trilinos using the 'jenkins' entity
account requires that all of the files being installed be group readable so
the install command can be run by another user in the group (the
atdm-devops-admin user in our case).  If the base build directory is propertly
protected with permissions this does not loosen the protection of this
'decomp' script at all (as it is just group readable not group writable).

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: MueLu: more work to reduce library size
PR Author: jhux2

Split this non-templated class into header and definition files.
Also switch to preferred default destructor for Monitor class.
Part of #4984.

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: SEACAS: cgns - fix zgc uniquification
PR Author: gsjaardema

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: Intrepid: Add cmake flag to make clear that Kokkos code is deprecated
PR Author: mperego

New xpetra eti attempt

Develop

This is modeled on MueLu's system.  I've converted
part of the class Xpetra::TpetraCrsMatrix.

This commit splits some MueLu classes into header and definition files.

Partially addresses #4984.

They can be deduced from the matrices..

Added a new paramter to the TimeStepControl, "Output Exactly On
Output Times", that controls if the timestep size will be adjusted
to exactly land on the output times for 'Variable' timestepping
(default=true).  When set to 'false' or for 'Constant' time stepping,
the timestep following the output time will be flagged for output,
and the timestep size will not be adjusted.

Modified Tempus_ExplicitRKTest::EmbeddedVanDerPol test to cover
new feature.  All tests pass.

 #4964