Align Functionspace.dofmap(s) interface

[schnellerhase]

Aligns Functionspace.dofmaps interface to geometry access patter, ref #4227. Otherwise current access patterns of mesh.geometry.dofmaps and functionspace.dofmaps do confusingly not align anymore.

• does not deprecate (python) FunctionSpace.dofmap, since also c++ layer still relies on it.

[chrisrichardson]

We should use .front and .at instead of []...

[schnellerhase]

I understand the .at. Why .front over .at(0)?

[jhale]

.front() is semantically correct from an STL perspective, works across all container types, and is cheaper (no bounds check), although this latter point is of course marginal in this non-hot loop setting.

[schnellerhase]

The question is whether we want to use .at(0) (bound-checked) or .front (not bound checked). Since every other access (and current usage) is bound checked through .at for consistency I think we should then stick to .at(0) unless there is another concern here?

[chrisrichardson]

We have used front elsewhere, so to be consistent I guess we should stick with that. It's not likely (or possible?) to have no dofmap.

[jhale]

As an aside, it's better/idiomatic to guard v.front() with !v.empty(), if needed.