Consolidate our usage of `hypre`

What is the requested feature?

The GEOSX validation cases basically use no feature of hypre.
But it’s our key linear algebra backend when it comes to big simulations.
This must evolve in order to provide a more reliable environment to our researchers and end-users.

Describe the solution you'd like

Here are some actions; feel free to extend or modify the list.

• Update geosxAts.
  Most probably by using a submodule (or some pip equivalent or anything else) to https://github.com/LLNL/ATS .
  This will prevent any further intertwining between our devs and the main framework. This will make future upgrades easier.
  Done in https://github.com/GEOSX/GEOSX/pull/2265
• Build integrated cases that use hypre. https://github.com/GEOS-DEV/GEOS/pull/4108
  • Majority of tests should have a direct solver variant, and an interactive solver variant.
• Build a policy (and tools) that validate results because the discrepancy of the result can be bigger than for direct solvers (due to the iterative nature of the hypre solvers). https://github.com/GEOS-DEV/GEOS/pull/4108
  • Criteria for iterative solver variant should be less exhaustive, such as norm of primary variable and iteration count information?? @castelletto1 @CusiniM please expand.
• Build “big” cases to validate the scalability.
  Can we rely on (semi) analytical cases?
  Do we have to play them as part of the integrated tests?
• Provide the hypre team with typical GEOSX validation matrix that could be directly embedded in the unit tests of hypre. @francoishamon
• Provide the hypre team with automatic checks to help improve the codebase of hypre.
• Create a repository containing our updated matrices, and get hypre tested against it.