Redesign of continuous and discrete fields

[biancagi]

This PR addresses two major issues in the previous design:

1. Class mito::fields::Field was a wrapper around mito::functions::Function, even though “field-ness” was already fully expressible as a type-level property of functions defined on coordinates.
2. There was a lack of symmetry between full-field analytic solutions to FEM boundary value problems and the discrete solution computed by the FEM solver (see for example Poisson benchmark). This asymmetry reflected in the fact that error norms could not be computed as L2 and H1 norms of the difference of the analytic and discrete fields.

To address the points above, this PR:

1. Removes the Field<F> wrapper and introduces the field_c concept to express “functions defined on a coordinate system” directly at the type level. Differential operators (grad, div, …) now operate on field_c functions, eliminating redundant wrappers and clarifying the separation between symbolic functions and geometric operations.
2. Removes asymmetries between exact and FE solutions in element-wise operations (e.g. error norms) and establishes a common, extensible localization interface.