Do not require get_form in user code

[ddundo]

Closes #223.

We only need the form when we do GoalOrientedMeshSeq.indicate_errors, so everything form-related was removed from MeshSeq and AdjointMeshSeq. Communicating the form from user code to GoalOrientedMeshSeq is now done with the read_forms method.

[acse-ej321]

@ddundo would it be worth adding a check in forms or _read_forms to check that the form has been passed by the user? I wonder if there is a good place in the class method templates or elsewhere to indicate this is needed for indicate_errors?

[ddundo]

@ddundo would it be worth adding a check in forms or _read_forms to check that the form has been passed by the user? I wonder if there is a good place in the class method templates or elsewhere to indicate this is needed for indicate_errors?

Thanks! I'll add this :) but the PR is still in draft phase so don't waste time looking into it now, as it might change :) I'll request a review soonish

[ddundo]

Ready now! Could you please review when you get the time? No rush at all.

Most changes are in demos, where stuff in get_form was just moved to get_solver. So it's a much smaller PR than it looks :)

[ddundo]

Updating forward and forward_old here only serves to update the coefficients in the form. Since there is no lagged solution in the variational form for "steady" fields, I do not update it here.

But the notion of a lagged solution still makes sense for "steady" fields in time-dependent problems. I kept this update for adjoint_next since in the next few lines we compute the average of it and adjoint. I'm not familiar with the theory here so I don't know if it makes sense or not, so I just left it as it is.

[ddundo]

(Unrelated to this issue :) )

[jwallwork23]

This is great @ddundo, thanks! Tidies things up very nicely.