Add support to save clustering statistics in lsstypes (and other) formats

[epaillas]

At the moment, the clustering estimator classes produce different kinds of outputs depending on the statistics (collection of correlation/power multipoles, coefficients, etc). Sometimes the data is saved as a simple 1D array (e.g. WST), sometimes as lists of pycorr objects (e.g. density split), etc, which can be confusing.

This PR begins to implement new save functions for the estimator classes, which can be used to save data as we currently are (for retrocompatibility), but more importantly, it allows us to save all statistics in a common lsstypes format (https://github.com/adematti/lsstypes), following the new standard for the clustering measurements produced by GQC. It also allows other formats (.nc or .zarr for xarray; only working for WST for now).

Output data format is given by the file suffix:

>>> wst.save('wst_coefficients.hdf5')  # lsstypes format
>>> wst.save('wst_coefficients.nc')     # xarray NetCDF
>>> wst.save('wst_coefficients.zarr')   # xarray Zarr
>>> wst.save('wst_coefficients.npy')    # numpy format

We can then read the data as e.g.

from lsstypes import read

data = read('ds_xiqg.h5')
xi = data.get(quantiles=4)
poles = xi.project(ells=[0, 2, 4])

[epaillas]

Ignore the "unit test for WST" commit, which is being reviewed in a separate PR.