Scenarios wrongly working when an ensemble of models is given as input

[amjjbonvin]

Describe the bug

In the protein-protein scenario, when running the standard protein-protein tutorial (or taking the input structures from the examples/docking-protein-protein haddock3 example) the web app generates very strange restraints to keep molecule 2 (B) together while there are no gaps in the structure. The example uses an ensemble of models. Something must go wrong in the restraints generation / break detection. This should be based only on the first model.

The problem probably comes from the PDB preprocessing that completely messes up the ensemble...
Compare the input ensemble from the haddock3 docking example (hpr_ensemble.pdb) with the corresponding ensemble generate as input for the workflow (download the inout archive)

As a result the docking results do not make any sense.

To Reproduce
Steps to reproduce the behavior:

1. Go to the protein-protein docking scenario
2. Input the PDB files from the haddock3 protein-protein docking example
3. Download the generated input archive and look at the preprocessed PDB file and restraints

Expected behavior

The preprocessed ensemble should be correct.

Additional context

This is probably affecting all scenarios in case an ensemble of models is provided.