Boundary conditions on nonconservative terms not supported

[jlchan]

Currently, boundary conditions are not imposed on non-conservative terms. See, for example UnstructuredMesh2D:

Trixi.jl/src/solvers/dgsem_unstructured/dg_2d.jl

Lines 426 to 432 in 7dd4232

	# Compute pointwise nonconservative numerical flux at the boundary.
	# In general, nonconservative fluxes can depend on both the contravariant
	# vectors (normal direction) at the current node and the averaged ones.
	# However, both are the same at watertight interfaces, so we pass the
	# `outward_direction` twice.
	# Note: This does not set any type of boundary condition for the nonconservative term
	noncons_flux = nonconservative_flux(u_inner, u_inner, outward_direction, outward_direction, equations)

The implementations for DGMulti and P4estMesh are similar.

[jlchan]

If we wanted the flexibility to set the BCs separately we would need to do something like pass a Tuple, e.g., with (boundary_condition_cons_part, boundary_condition_noncons_part).

Passing in a tuple for specialization makes sense! It'd be easy to specialize BoundaryConditionDirichlet, BoundaryConditionDoNothing, and even custom boundary conditions for this. I think this might result in a breaking change though, so we'd have to save it for 0.6.

#1431 (comment)

[sloede]

Even though it's a breaking change, it should not be a braking change. If somebody needs and implements this, we will happily bump the minor version to accommodate this.

[jlchan]

Well done with the pun

[ranocha]

Including this, there are at least two things we would like to change soon - the other one is your breaking change to the parabolic stuff. We can just do it soon, I think.