MG performance upgrades

[lroberts36]

PR Summary

This PR makes some changes to make things go faster, including:

1. Caches PackDescriptors for boundary conditions generic using a map. At least in MG, re-creating pack descriptors every call was a significant performance hit.
2. Re-arranges internal fields created by solvers to minimize the number of fields that share ghost data.
3. Adds mesh data for MG communication that only includes internal fields (so that fields from user packages are not communicated unnecessarily during MG solves).
4. Adds hierarchical parallelism in some of the MG utilities.
5. Adds a setup phase for MG to reduce the number of fields that are restricted.
6. Adds some solver parameters.

This breaks downstream codes that use a solver, since now solvers require a call to AddSetupTasks before calling AddTasks.

PR Checklist

• Code passes cpplint
• Code is formatted
• Changes are summarized in CHANGELOG.md
• CI has been triggered on Darwin for performance regression tests.
• (@lanl.gov employees) Update copyright on changed files

[Yurlungur]

I think it would be useful to better document the solver parameters and how to use the solver more generically. But overall this LGTM.

[lroberts36]

I think it would be useful to better document the solver parameters and how to use the solver more generically.

Agreed, but I am not sure that the dust has settled on the solver stuff yet. I mostly want to get these changes in now so the branch doesn't go stale.

[pgrete]

Just a couple of clarifying questions, so feel free to merge when ready.

Otherwise, I second @Yurlungur comment on documentation ;)