Change the default value of the `depth` argument in `EquivalentSources`
to `"default"`. If this value is passed, then the depth of the sources
is set as 4.5 times the median distance between first neighbouring
sources. If a numerical value is passed, this is the one that will be
used. Introduce a new `depth_` attribute where the estimated/passed
numeric value is stored. If `points` is passed, the `depth_` attribute
is set to None. Add a test checking that the depth value is properly
set.

Merge the tests functions that checked the accuracy of equivalent
sources using different dtypes.

Add admonition explaining the default behaviour of the ``depth``
argument.

Use the mean to compute the average distance of neighboring data points.

Try to make tests on Python 3.8 pass

Co-authored-by: Leonardo Uieda <leo@uieda.com>