Better BC signaling

[andrewreisner]

The current boundary changed signal requires the operator to be reassembled and MG setup phase to be rerun. For a Schwarz solve, it would be useful to have a signal for a change to the Dirichlet boundary values. This would only require reprocessing the right-hand side resulting in a significantly more efficient Schwarz iteration.