Applying nan_to_num for Effective Area

[chaimain]

Effective Area evaluation is missing the application of numpy.nan_to_num in the functions effective_area_per_energy and effective_area_per_energy_and_fov in https://github.com/cta-observatory/pyirf/blob/main/pyirf/irf/effective_area.py unlike for other IRFs.

[maxnoe]

It's right that this is inconsistent, but I'd argue we should remove the nan_to_num from the other IRFs instead of adding it to effective area.

The nans occur when dividing by zero, which always means "no true events in this energy bin". So we just don't know the IRF, filling nan is more appropriate than filling 0.