- Keep track of the membrane facing atoms that contributed to the energy
  function and map their energies as float properties to the input entity
- Don't consider atoms with energy contributions of 0.0 in the optimizer to
  to speed things up

When exporting the scene for the second time, weird artefacts that looked
like shadows occured. The fix enforces to re-allocate the resources of
the offscreen QOpenGLContext. Measurements showed that reallocation is
much faster than the rate-limiting step of the Export function: dump the
file to disk. Performance impact is therefore limited. Feel free to
investigate and propose a more elegant fix...

This bug is triggered only when mappings have the exact same RMSD and
python tries to inspect the dictionary for ties.
Added test for this specific case.