CONECT records cannot be parsed for systems with >100k atoms

[hannahbaumann]

Tools like maestro switch to hexadecimal outputs after 100k atoms, e.g. A000G.
CONECT records in this format raise an error in the PDBStructure loader.
This PR #663 adds support for parsing these.

This is how mdtraj re-wrote pdbstructure.py to parse the hybrid36 convention:
https://github.com/mdtraj/mdtraj/pull/1849/changes

MDAnalysis issue discussing this: MDAnalysis/mdanalysis#5089