FEniCS backend: assignment and interpolation updates

[jrmaddison]

The FEniCS backend previously handled Function.assign([linear combination of functions]) using the ExprInterpolation class. However this seems fragile in the case of non-continuous functions. This PR removes this case. A better approach for non-variable coefficients is to use LinearCombination directly, otherwise a custom Equation class will now be needed.

The FEniCS backend interpolate_expression now always uses point evaluation via Expr for primal Function interpolation. This is slow, but I'm not aware of a good general solution with legacy FEniCS. This may also change the behavior for DG functions, but interpolation is ambiguous anyway in this case.

Firedrake note: The Firedrake backend version of ExprInterpolation uses Firedrake's much faster approach for interpolation of general expressions. The Firedrake ExprInterpolation is still used to transfer functions from restricted to non-restricted spaces, even for non-continuous spaces.

Also:

• FEniCS backend: Catch the case Function.assign([function in a different space]) (not caught by legacy FEniCS).
• Update some tests.