Fix energy_resolution method in EnergyDispersion class

[lmohrmann]

The energy_resolution method in EnergyDispersion could only handle 1-dimensional inputs so far (see also #1132). With this PR, N-dimensional inputs are allowed:

import numpy as np
import astropy.units as u
from gammapy.irf import EnergyDispersion
e_true = np.logspace(-1, 2, 30) * u.TeV
e_reco = np.logspace(-2, 2, 20) * u.TeV
edisp = EnergyDispersion.from_gauss(e_true, e_reco)
print(edisp.get_resolution([[10, 20],[30, 40]] * u.TeV))
print(edisp.get_resolution(10*u.TeV))
print(edisp.get_resolution(20*u.TeV))
print(edisp.get_resolution(30*u.TeV))
print(edisp.get_resolution(40*u.TeV))

[[ 0.20902851  0.20850442]
 [ 0.20678505  0.19102049]]
0.209028510522
0.208504420638
0.206785053106
0.191020494988

[joleroi]

Thanks @lmohrmann
In the end, it would be nice to have tests for all these functions, maybe you can also have a look at #1122

Regarding this PR is does not seem to break anything, so go ahead if you'd like to merge it

[joleroi]

If I would have to moan about something, I'd ask you to put some comments here, what all this is doing. It's kind of annoying (at least for me) to have to decipher numpy expression all the time 😄

[lmohrmann]

Yeah, this is a bit ugly, but I guess this can hardly be avoided if N-dim arrays are supposed to work. The documentation I just added should make things a bit clearer, I hope.

[lmohrmann]

Merging this now, although I agree that the tests should be improved. This is better done in line with #1122, I guess.