fixes the issue of a wrong PSF normalization. This issue has been seen in the Crab validation as a mismatch of the residual plots in the following two (apparent equal cases): Use the PSF distribution + point-like source model fit vs. not relying on the PSF + assuming a gaussian source model.
To fix this we now divide in theta^2 bins and normalize to get dP/dOmega. The integral over the energy axis (2 np.pi rad_width_deg normed (rLow + rHigh) / 2).cumsum().max()) in units of deg^2/sr should be equal to (180/pi)^2 ≈ 3282.8
add lower and upper energy axis limits
add possibility to generate fake PSF using a halfnorm distribution to test the pipeline (converter -> gammapy) or for cases one wants to overwrite it for some analysis cases