Simplify the __init__ of the MyScaler class

[pierrePalud]

Currently, the __init__ method of the MyScaler class is written to either

• take a grid in input and use it to compute the mean_ and std_ vectors (used in the normalization) -- this approach currently raises an NotImplementedError().
• take directly the mean_ and std_ vectors in input.

In the rest of the code, the mean_ and std_ are in practice loaded from a sklearn.preprocessing.StandardScaler() object and then given to initialize the MyScaler instance, which seems needlessly complicated.

To clarify the code, I suggest:

• to remove the option of providing a grid to the __init__ (which in any case is currently not supported)

[pierrePalud]

I removed the option, as described above.

In addition, to avoid handling the scaling factor kappa as a special case in _from_scaled_to_lin and _from_lin_to_scaled, I changed the self.mean_ and self.std_ numpy arrays, such that if there is a kappa in the physical parameters, it should be sampled in log scale. Thus I set

• its mean to 0
• its std to 1 / np.log(10).

This value of the std was chosen for the common validity interval on kappa, [0.1, 10], such that in scaled space it becomes [-e, e], that is, approximately [-2.7, 2.7]. This scaled interval has a similar size as the scaled validity intervals for the other physical parameters (that are all normalized).