Implement HDivTrace as a macroelement

[pbrubeck]

This PR enables macroelements defined on subcomplexes of the reference complex.

As a result we can reimplement HDivTrace as a nodal CiarletElement

HDivTrace(ref_el, k) := DiscontinuousLagrange(TraceSimplicialComplex(ref_el), k)

[rckirby]

This seems very pervasive, and not clear sometimes what's cleanup versus what's needed for the functionality. Were generalizations (say, to Lagrange) really needed to support this? Maybe it's not surprising if trying to do H(div) traces as a macro element spawns a bunch of problems way down the chain that we hadn't thought of.

[pbrubeck]

Passing Firedrake CI firedrakeproject/firedrake#4429

[pbrubeck]

This seems very pervasive, and not clear sometimes what's cleanup versus what's needed for the functionality. Were generalizations (say, to Lagrange) really needed to support this? Maybe it's not surprising if trying to do H(div) traces as a macro element spawns a bunch of problems way down the chain that we hadn't thought of.

I agree, this required to enhance ExpansionSet and LagrangeExpansionSet (the barycentric interpolation code, not the CiarletElement) to deal with codimension 1 (or higher) facets.

I think regardless on whether or not we want to reimplement HDiv Trace as a macroelement, these enhacements should be kept.