Get ride of Connectivity::invCellNbNodes()

This was a marginal optimization that
- is too specific (was used only once ...)
- generate no measurable performance change