Vertex based slope limiter is overly diffuse near boundaries

[Stephan Kramer]

This is a long standing issues with the vertex based slope limiter, used both for momentum dg and dg scalar advection. Because the limiting is only based on cell averages any gradient in the cells directly adjacent to the boundary are immediately flattened out. This is an issue now in a lot of cases, e.g. temperature stratification is immediately flattened near the bottom and top, and flow in a channel with free slip bcs acquires an artificial boundary layer.

Proposed solutions:
* for cases where a Dirichlet bc is applied include the bc value in the limiting values (sparkinson)
* for cases with no such bc, compute facet averages and limit with those (skramer) - this needs a bit more thought and evaluation whether this is actually proper and correct