Change energy dispersion format

[cdeil]

This is a suggestion for a change I got via email from Michael Punch and @kosack, after presenting the FITS data formats for H.E.S.S. at the collaboration meeting last week.

There we were using EDISP in the current format (see here) with y-axis range up to y = e_reco / e_true = 2 which clipped the distribution.

To avoid this they suggested to instead use y = log(e_reco), which is what OGIP RMF does, and also use that for the DL3 responses (where there are extra axes e.g. for field of view offset, and no sparse matrix zero-suppression scheme).