Catch the numpy LinAlgError when recostructing the stereo parameters.

[aleberti]

Closes #125

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (e4b9908) 76.83% compared to head (1cec705) 76.83%.
Report is 6 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #189   +/-   ##
=======================================
  Coverage   76.83%   76.83%           
=======================================
  Files          21       21           
  Lines        2556     2556           
=======================================
  Hits         1964     1964           
  Misses        592      592           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jsitarek]

so if you catch such an event with a problem it will skip the calculation of remaining stereo parameters and saving them into the output for this event, therefore the Hillas parameters will be still there, but stereo parameters will have 'NaN'. This could potentially cause some issues later on in the code (e.g. when filling in data for RF), but the same was applied before for non valid stereoscopic reconstruction and I checked in recent files that I had such NaNs, but no problems in RF stage, but I guess they are automatically removed there. So perfectly fine the way how you did it.