Olm is not a proper framwork for FiDo

This is a meta bug since it is design flaw in the current "implementation".
The framework for Olm has been (ab)used to demonstrate some Nitsche-based fictitious domain formulation.
But recent examples have proven that approach to be inefficient and error-prone.
Main points are:
-Olm requires two full 3d meshes whereas Fido needs only a surface.
-Faking a surface by using the boundary of a second 3d mesh is highly inefficient for collision detection.
-When the boundary of the second mesh has the opposite orientation of overlapping mesh expecting in OLM,
many things go wrong, ranging from the proper cut cell computation (computes the complement of the cut cell)
and mesh marking (switch between completely and not overlapped mesh cells).

Fido needs to be sourced out.