Don't pass restricted BCs into assemble

[leo-collins]

Don't pass the restricted BCs in to assembly. When restrict=True, the boundary DoFs are removed in the restricted function space so passing in these BCs does nothing. This lets us use restrict=True for residuals containing things we cannot apply BCs to (e.g. cross-mesh interpolation currently) in the traditional way.

[pbrubeck]

We still needs the BCs for a RestrictedFunctionSpace, as they are used to set the value of u_restrict at the beginning of the solve. Does your change preserve that?

[pbrubeck]

Why is this change required? Please add a description and or tests.

[leo-collins]

We still needs the BCs for a RestrictedFunctionSpace, as they are used to set the value of u_restrict at the beginning of the solve. Does your change preserve that?

I only change the BCs passed to the residual and Jacobian assembly. problem.bcs is unchanged.

Why is this change required?

Because we don't need to pass these BCs in to assembly. In particular I would like this for my cross-mesh stuff, since BCs for cross-mesh interpolation has not been implemented yet. I will add a description and tests.

[connorjward]

I hit this with pyop3. It was quite annoying to have the debug why imposing boundary conditions on a restricted space failed to be a no-op.

More generally I don't understand why it is valid to create a DirichletBC object on a restricted function space. I think it just leads to confusing corner cases. For instance what if the boundary_set between the BC and function space differ?