SURs do not hold things such as normals, faces, etc. They’re strictly there to act as collision detection. The SURs are also noticeably simpler in their mesh than CMPs are, otherwise collision detection would take far too many CPU cycles to process (At least, back when Freelancer was released).
SURs and CMPs are two completely different formats, as far as I know. They probably have similarities, since they carry similar information, but I’m not versed enough into either formats to provide a better explanation.
As for the last question, that’s a definite no. CMPs without SURs have no collision detection, will often disappear once their center gets out of the field of view (effectively making the whole model pop in or out), and you’ll have a hard time selecting those meshes because the game also uses SURs to define where is a mesh selectable.