Fix vertex mesh compilation with geometry tables.

[jorgensd]

An issue I was already aware of due to work in Dolfinx-adjoint and FEniCSx_ii (reported by @finsberg).
Evaluating spatial coordinates on a point mesh has been broken for a while in FFCx.
This was also reported in: #879

The issue in the existing code is that the coordinate element of a point mesh is a DG-0 function.
This returns a basis table of a single number (1), which is correctly classified in the table generator as ttype="ones".
However, in FFCx we drop all "ones" and "zeros" tables, and therefore have to manually handle code generation for those cases.

This PR fixes this for coordinate dofs. The new generated code becomes:

double x_c0 = coordinate_dofs[0];
double x_c1 = coordinate_dofs[1];
double x_c2 = coordinate_dofs[2];

for a 3D point.

AI disclosure

The test was manually implemented by me.
The fix was created with the help of CLAUDE (Opus 5). The code has been revised and simplified by me.