Interaction of trace space and reference states in linear solver

[Thomas Bendall]

The reference states need manipulating in the linear solver, by averaging to the
trace space. Previously this was done in the initialisation of the linear solver, but now the reference states don't exist when the solver is initialised. In theory, these reference states could evolve with time, so the calls to the averaging operator are now done every time the linear solver is actually called. However this is normally very inefficient -- could be move the averaging operations to their own routine and keep track of whether or not they've been initially called, and have an option for time-updating reference states?