Fix Stokes velocity FE dual-space setup for high-order simplex solves

[gthyagi]

Summary

This fixes a Stokes FE setup inconsistency that breaks higher-order triangular velocity solves.

In SNES_Stokes_SaddlePt._setup_discretisation(), the private velocity PetscFE was created with the requested polynomial degree, but without the matching PETSc dual-space options that UW3 already uses for mesh variables and the other solver setup paths.

This PR adds the missing velocity FE options before PETSc.FE().createDefault(...):

• private_<prefix>_u_petscdualspace_lagrange_continuity = self.u.continuous
• private_<prefix>_u_petscdualspace_lagrange_node_endpoints = False

Reproducer

A minimal reproducer script is included in this PR:

• examples/stokes_box_mms_simplex.py

Run it directly from the repo root, for example:

python examples/stokes_box_mms_simplex.py
python examples/stokes_box_mms_simplex.py --vdegree 2 --pdegree 1
python examples/stokes_box_mms_simplex.py --vdegree 3 --pdegree 2

The script uses a simple manufactured Stokes solution on the unit square with triangular elements:

• stream function: psi = x^2 y^2
• exact velocity: u = (2 x^2 y, -2 x y^2)
• exact pressure: p = x^2 - 1/3
• exact Dirichlet velocity is imposed on all four sides

This is a clean test because the exact velocity is cubic and the exact pressure is quadratic, so P3/P2 should represent the solution essentially exactly on a straight box.

Error norms before and after

Using the same box MMS setup at h = 0.125:

Before this fix:

• P2/P1: velocity 1.5068686519735996e-04, pressure 3.9062499996873525e-03
• P3/P2: velocity 1.2277197041312973e-02, pressure 2.0668703128596483e-10

After this fix:

• P2/P1: velocity 1.5068686519735996e-04, pressure 3.9062499996873525e-03
• P3/P2: velocity 3.6635897127143294e-12, pressure 1.796728310502028e-10

So the fix leaves P2/P1 unchanged and restores P3/P2 to the expected near-machine-precision result on this exact polynomial test.

Notes

This PR is intentionally minimal:

• one solver-side FE setup fix
• one small reproducer script

It does not include the separate imported-mesh / XDMF work that came up during debugging.

[lmoresi]

Reviewed the fix and audited all PETSc.FE().createDefault() call sites across the codebase.

Stokes velocity fix (Tyagi's original commit): Correct and minimal. The Stokes velocity FE was the only solver field missing the dual-space options — Scalar solver, Vector solver, and the Stokes pressure FE all had them already.

Added in follow-up commit:

• Comment explaining why these options matter (node placement on simplices, P2 works by accident, P3+ breaks)
• Same fix applied to the mesh coordinate projection FE in discretisation_mesh.py — same omission, currently harmless for P1 coordinate meshes but would break higher-order (curved) coordinate meshes on simplices

All other FE creation sites (MeshVariable, gradient evaluation, field projection) already set the dual-space options correctly.