The following issues were observed:
- warning about non-writable directory which was set as IPYTHONDIR env var
  Not very surprising, it's a directory that lives in container space...
- stale kernels
  I didn't investigate that one properly. It had something to do with bending
  around all Jupyter/IPython relevant directories to custom set directories
  in the image. Reason for doing that is potential Python modules that reside
  in user home of the host that get imported in the container. I removed the
  bending around with the full knowledge that the described problem may occur.
- version updates
  Some changes related to different Python versions between old base image
  (Ubuntu 20.04) and the new one (Ubuntu 22.04).

This makes the ligand scoring script more independent of an up-to-date
compound library, which is a strong requirement of Molck.
This was a problem for the April 2023 Modeling Championship where the
script was run with an out of date library and the element of an atom of
the ligand was silently changed, preventing the symmetry correction to
find isomorphisms and resulting in the absence of scores.

OST writes invalid atom serial numbers if they exceed the PDB format limits.
That's a feature not a bug. The new behaviour is that invalid serial numbers
don't get parsed. Any processing function working with these serial numbers
need to manually check for that. The only current effect is when reading
CONECT statements. Atoms with invalid serial numbers are not connected.

Required to build OST on latest sciCORE toolchain. Already tested
but forgot to commit. Not sure how Inaki managed the RC modules.

Behavior was made more consistent. Methods are now available
on all the (Atom|Residue|Chain)(Handle|View)s, including in
Python, and documented. A hash code is an `unsigned long`,
and requesting it on an invalid handle raises an `InvalidHande`
exception.
In practice, the following was implemented:
- Exported to Python for ChainHandles
- Added method on ResidueViews and ChainViews (including Python export)
- Raise InvalidHandle error on ResidueHandle (was: return 0)
- Explicitly check validity of ChainHandle (although behavior was to
  throw an error even without explicit check)
- Changed return value from `long` to `unsigned long` on
  Atom(View|Handle)s.

GetPrev() or GetNext() may return invalid residue handles. It is
unclear what happens next, find() may simply (hopefully) find nothing,
however this feels unsafe. This fix guards against this possibility
by checking for handles validity before giving them to find().

Simon found issues when reading and writing CONECT statements
in PDB io and provided code to fix it. yay

Reading: OpenStructure did not read CONECT statements at all. By default,
it still doesnt. Many PDB files out there don't provide these
statements and we really don't want to rely on them. We rather want
to assign connectivity based on the chemical component dictionary from
the PDB. However, a valid use case are novel compounds that are not
in the component dictionary. Reading connect statements can now be
enabled in the pdb reader through the IOProfile. This may give issues
in processing after reading. OpenStructure implements processors that
are responsible for connectivity. Now that we build some of the
connectivity already at the reading stage, this might cause trouble.
To remedy most of the nightmares that can come out of that, the
processors can now optionally skip connectivities between Hetatoms.

Writing: That was a plain bug when writing CONECT statements for
bond orders > 1.

PDB announced they're running out of three letter codes for compounds and
switch to longer names soon. This commit ensures that OpenStructure is ready
for that switch.

This may be necessary in case SDF ligands have the same chain/residue
number, which can happen with some input.

This should clarify what the binding site.