Revise cell/facet size quantities in all DG demos

[Jan Blechta]

There are sketchy quantities used in DG demos like simplex circumradii. Let's do the correct mathematics. This depends on solving #924.

[Jan Blechta]

This includes DG demos in DOLFIN and FFC.

Note attempts of making DG demos use quads/hexes as in pull request #381. It should be ensured that after this revision the switch to quads is possible (and can be eventually done at the same time).

[Jan Blechta]

See #664 too.

@dnarnold, is cell (facet) diameter (diameter of a set in mathematical set) a preferred quantity in DG forms rather than cell (facet) circumradii?

[Douglas Arnold]

Yes, I believe that the cell or facet diameter is the commonly used quantity in DG. For example, in the coefficient of the penalty term in an interior penalty method for the Poisson equation there is a factor of 1/h^2 and h is most often taken as the diameter of the facet containing whose jump is being penalized. The circumradii of the facet would bring into account the shape as well as the size of the facet, which, to my knowledge, is not common or theoretically supported.

[Jan Blechta]

Ok, we have now new CellDiameter instead CellSize=2*Circumradius in all DOLFIN demos except of multimesh demos (which do not run with uflacs).

FFC demos were not touched. They are only important for regression testing. Much broader coverage for (new and old) geometric quantities added to test/unit/python/fem/test_symbolic_geometry_assembly.py:test_geometric_quantities.