Support different map energy binnings in MapMaker

This is a follow-up issue for #1579 and #1580 by @AtreyeeS that weren’t merged.

The question is whether to support different energy binnings for different maps in MapMaker.

Currently all maps (counts, exposure, background) have the same energy binning. The counts and exposure have a reco energy axis, and the exposure has a true energy axis, exposure combined with EDISP is not supported yet. Since #1558 we integrate background in the power-law approximation, using nodes at the energy bin boundaries (this number of nodes option not re-exposed on the Mapmaker, which I think is OK); for exposure we just evaluate at the energy bin centers, the user has no way to control precision except to choose not too large energy bins.

I’m not sure if it will be commonly used, but I’m +1 to add a energy_true=None config argument to MapMaker. By default, energy_true = energy_reco is used and energy_reco is taken from the geom (which defines the counts geom, with a reco energy axis). If the user wants a different energy_true, they can pass a quantity. That’s pretty similar to how the user can choose any energy_true and energy_reco binning they like in gammapy.spectrum.SpectrumExtraction.

@AtreyeeS @registerrier @leajouvin @adonath - Thoughts?

[registerrier]

An energy_true vector/axis is clearly needed to create the relevant maps.

Note BTW that a different MapGeom for energy_true could be a valid possibility too. Because of ROI sizes and contamination, one might be forced to model the true sky on a larger space region than the observed one.

Also, for computation efficiency, it is obvious we don’t need to compute reprojected IRFs on a very fine binning (e.g. 0.01°) when the IRF are sampled on a grid 10-50 times less precise. It would be MUCH more efficient to compute exposure and background maps on maps 10 times less precise than the counts maps and at the end of the process interpolate the average map to sample it on the final binning. We might therefore also pass an oversample argument to the MapMaker to compute the reduced IRFs.

[AtreyeeS]

Done in #1846