Coloring and map_element_nodes do not play well together

[Andlon]

This applies to VectorParAssembler and CsrParAssembler, both of which require coloring. Usually coloring happens at an earlier stage than when you'd use map_element_nodes, meaning that after map_element_nodes, the coloring no longer corresponds to the mapped indices.

I'm not sure how best to fix this in the short term - at the very least, the assemblers should detect this issue and return an error, which they currently don't do. Long term, we anyway want to move to parallel assembly that doesn't longer coloring, which should resolve this issue that way.