Draft: AMR preliminaries

[hughcars]

Description of changes:
Minor changes in advance of the AMR PRs:

• Some constexpr
• Fix some enum and unused variable warnings
• Broke out the default integration rules into a function, this was required by an initial implementation, but no longer necessary. Included however as it was a nice simplification from multiple inheritance.
• Refactor the multgrid helper function signatures to use overloads, and allow building p-multigrid or single mesh/fec "multigrids".
• Added GlobalTrueVSize() to main operator classes, necessary for future estimation PR.

Issue #, if available:
#2

[hughcars]

Changed the base to sjg/libceed-pa-dev in anticipation of merges.

[sebastiangrimberg]

Changed the base to sjg/libceed-pa-dev in anticipation of merges.

To me the changes to integrator.hpp, multigrid.hpp, and main.cpp all seem unnecessary and don't really provide any added value. At that point you might as well just fold this into your first AMR PR, unless it makes sense to keep the meshio.cpp constexpr changes separate in which case that's fine too to merge in ASAP.

[hughcars]

Changed the base to sjg/libceed-pa-dev in anticipation of merges.

To me the changes to integrator.hpp, multigrid.hpp, and main.cpp all seem unnecessary and don't really provide any added value. At that point you might as well just fold this into your first AMR PR, unless it makes sense to keep the meshio.cpp constexpr changes separate in which case that's fine too to merge in ASAP.

These were mainly the small style type changes I did whilst doing exploratory work within the amr-dev branch, but they weren't really related to the actual prs coming. The approach we've had in previous PRs is to aim for smallness and to avoid any extras, so I broke them out of the main PR sequence in anticipation.

The only thing that's truly needed from here is that NDof, which I will change to GlobalTrueVSize, the rest are unrelated to the ultimate PR sequence but I thought were slight improvements:

• The constexpr is more idiomatic c++.
• The immediately invoked lambda in main.cpp allows for binding a const variable.
• Namespace over static method is also more idiomatic c++. Moving the quad rule into a function was helpful in an early iteration of the estimation algorithm, where I had been making a function that built the element integral operator on the fly. I figured it expands the capability of getting the rules without needing inheritance. Basically, that it can be done with namespace rather than a static method means it ought to be.
• The multigrid constructor methods makes building without boundary conditions more obvious (at least to me) compared to passing the nullptr. This makes the call sites in the estimation class look neater too, with the dispatch being moved to compile time, rather than on the runtime status of the pointers within the body. The templated template also has the nice property it plays nice with passing in a unique_ptr or with passing a vector of unique ptrs, which is nice if we're getting rid of the grid hierarchy as h multigrid goes away (I had a version of the main helpers that only did P-multigrid for instance, and used the essential boundary function). I considered using default arguments of nullptr, but that doesn't help if dim then needs to be specified.

[hughcars]

Being rolled into other PRs.