fix WV x-spectra stacking

[agrouaze]

recommendations (F.Nouguier):

1. find the smallest size in range and azimuth for all the xpsectra WV dataset (2millions imagettes)
2. crop all the spectra (to minimal values found) using isel
3. store the k_rg and k_az as variables in the stacked dataset
4. look at the variability of the 2 vectors (k_az and k_rg) in the dataset to decide if they are close enough to be considered as identical. If not -> find the smallest k that are "common" and crop them.

My current approach is more brutal, I crop the xspectra to k<0.14 (40m) and it works (so far) because the data is close enough at these values of k.

This PR also contains modifications to allow to build the dataset WV L1C stacked containing groups for foundation model training:

group: ecmwf_0100_1h {
group: ww3_global_yearly_3h {
group: sar_sentinel1_wv_l1b_xsp_intraburst {