New features

Add a warning to xoa.open_data_sample that is emitted when the request edfile
is not an internal data sample [#47].
Add the xoa.plot.add_shadow, xoa.plot.add_glow and
xoa.plot.add_lightshading function to add path effects to plots [#44].
Add the xoa.plot.plot_ts function to make T-S diagrams [#43].
Add the xoa.filter.demerliac function to apply a Dermerliac filter
to time serie [#41].
Add support for fine tuning masking in xoa.filter.convolve through the na_thres
parameter [#41].
Add the xoa.geo.cdist and xoa.geo.pdist functions to compute
haversine distances respectively between two dataset and with a dataset [#40].
Add the xoa.coords.geo_stack function to stack longitudes and latitudes
into another dimension, in a dataset or data array [#40].
Add the xoa.filter.decimate function to crudely undersample a geographic
dataset or data array with a radius of proximity [#40].
Add the xoa.geo.get_distances to compute the Haversine distances between
locations inside a single dataset or between txo datasets [#40].
Add the xoa.krig.empirical_variogram function to estimate variogram parameters.
Add the xoa.krig.VariogramModel to manage a variogram model [#40].
Add the xoa.krig.Kriger and xoa.krig.krig to perform kriging [#40].
Add the exclude option to data var and coordinate specifications of
xoa.cf.CFSpecs instances [#38].

Breaking changes

A single Nan now contaminates the data over the kernel emprise in xoa.filter.convolve
since na_thres is set to zero by default [#40].
xoa now requires the gsw package.