Bug fix for overlap_sort() modifying self

[dbochkov-flexcompute]

No description provided.

[momchil-flex]

My general feeling is that it should be possible to simplify some parts of the code using xarray directly to avoid things like np.exp(-1j * phase[None, None, None, :, :]) and

                field_sorted.data[..., freq_id, :] = field_sorted.data[
                    ..., freq_id, sorting[freq_id, :]
                ]

One of the big advantages of xarray is that you don't have to remember which array axis a given coordinate corresponds to (and if axes get reshuffled for some reason later on, code will still work). This is a comment not just on this PR but on the sorting code as a whole, which seems to use raw numpy a lot. However, changing this would require some thought (despite all its benefits, xarray is still not super intuitive to me) and seems like a pretty low reward / effort ratio, so I'm just leaving this comment here for the future but otherwise PR looks good!