Incorrect wall distance using axisymmetric mesh

On "axysymmetric" cases (where the mesh uses a radially extruded slice, with symmetry on the sides), the wall distance approximated by a diffusion equation (in ``distpr``) can be imprecise, with a difference up to 40% having been observed.

This issue also occurs using the CDO operators, so seems due to the the choice of equation or boundary conditions rather than linear operator.

In any case, this can induce an error on turbulence variables.

More works needs to be done on this, but replacing the approximate wall distance with a
geometrically computed wall distance for the boundary cells might at least mitigate the effects.

Attached is a small case allowing to test this.
[test_pipe_axi.tar.gz](https://github.com/code-saturne/code_saturne/files/3135105/test_pipe_axi.tar.gz)

Start with the attached test_pipe_axi.tar.gz case and the distpr wall-distance calculation; compare its axisymmetric result with the CDO-operator behavior. Investigate whether the equation or boundary conditions cause the reported 40% difference, and assess the proposed geometrically computed distance for boundary cells. Done means the test case no longer shows the problematic wall distance or its turbulence-variable error.

Written by the indexing model from the issue text.