SNR calculation

[auggiemarignier]

In darkmappy.validation the snr is calculated as

return 10.0 * np.log(l2_true / l2_diff)

but it should use
return 20.0 * np.log10(l2_true / l2_diff)

Need to be careful with the np.linalg.norm(x, ord=2) calls to calculate L2 norms as this behaves slightly differently for arrays (images) or flattened vectors. Also it doesn't return the squared norm, hence the new factor of 2.