[surface model] fix conical deduplication
The conical deduplication was incorrect for surfaces with a global transformation. The idea was to compare the radius at 0 in the global reference frame, the minus in the implementation was incorrect.
The problem was identified and confirmed as resolved with a part of the CMS2018 geometry.
Prior to this fix, this could trigger infinite loops where the navigator was stepping by some very small distance before doing the same small negative step.