Provide FITS file implementation of the `CustomInstrument` classes

[sguillot]

The CustomInstrument classes provided either in the tutorials or the Zenodo can be confusing. I think it could be greatly simplified by allowing the use to load directly from the FITS files. Response matrices (RMF and ARF) are standardised in X-ray astronomy, so there should not be any complications. It should be also relatively easy to extract the channels edges and to figure out from the file header if the RMF has already been multipled by the ARF or not.

This suggested feature should probably be tackled after the python3 release (since reading FITS file will rely on Astropy, and we will likely face difficulties with astropy version that are only compatible with Python 2)