Handle production of NaN during angle calculations.

[HubLot]

Numpy can produce NaN during dihedral calculations (in get_dihedral function). This can occurs when we try to normalize the vectors ("Divide by Zero"). It usually means there is issues with some residus coordinates (for example, 2 residus superposed). The problem is the dictionary of phi/psi angle is still returned and the assignation is still made despite the NaN in the angle. This produces incorrect assignations which could be difficult to spot on a analysis of many PDBs.

This PR fix those issues by :

• Transform the RuntimeWarning produced by numpy in a proper FloatingPointException exception (through numpy.errstate(invalid='raise')). This change is made in get_phi_psi_angles() function before the main loop on the residus. Hence, it's only done once and do not change the speed of the calculations (it's even a little bit faster but not sure it's significative).
• Re-raise the exception in get_phi_psi_angles() function in order to keep the traceback and provide a proper exiting tool for the function (helpful for module users).
• Catch the exception in PBassign, warn the user about the issue and don't write the assignation.

[pierrepo]

Nice work @HubLot

[HubLot]

shit I forgot the CHANGELOG :(

[HubLot]

@pierrepo, can you commit directly to the master (bad I know) with only the modification in the CHANGELOG file (like 'Handle production of NaN during angle calculations') ?

[pierrepo]

@HubLot already done + reformat comments

[HubLot]

great, thank you!