Don't have dt as an argument to the physics evaluate methods

[tommbendall]

Some physics schemes need to know the time step (separately from the time discretisation).

In case the time step changes, we currently pass dt into every evaluate method. It would be better to have physics.dt point to time_discretisation.dt, and set this in some set up method