Custom C/C++ kernel demo

[jhale]

Work towards solving #2942.

First draft for discussion. P1 Mass matrix assemble.

[jorgensd]

Blank slate to work towards solving #2942.

First initial thought is (and this applies to the Python interface too); how can I make a function space with dofs assigned to vertices without codegen?

@jhale see; https://github.com/FEniCS/dolfinx/blob/main/cpp/demo/interpolation-io/main.cpp#L34-L42

[jorgensd]

@jhale
Is there a reason that our custom kernel has to use pre-defined finite element tabulation?
I think it would be really nice to show that you can make up your own quadrature scheme (or take one from a text-book), and use basix to do runtime tabulation.

[jhale]

No, it's just a placeholder. I plan to add tabulation via Basix.

[jhale]

Should we assembler for an FFCx version too to compare the result? This would also be super helpful for profiling the kernel call cost (since the pure C++ version can inline to call).

Let's discuss this on Slack in #development - I don't think this is currently possible as std::function is everywhere. If we can get that issue solved, yes, I'd be happy to produce a counterpart for benchmarking/comparison.