The number of features should have no influence
on the scalability.

Add querying for KNN

See merge request jjerphan/cython_plus_experiments!1

This now uses this fork:
https://gitlab.inria.fr/jjerphan/cython

This reverts commit 27426854.

Reverting to experiment with the original runtime again,
to see if removing the logging helps.

This reverts commit 39363f1c.

Reverting to experiment with the original runtime again,
to see if removing the logging helps.

The target commit removes the logging on contention
which slows the execution of the code.

Round robin is not thread safe.

It would require putting an atomic on ``next_worker``
which would probably slow dispatching queues.

Hence prefering random dispath.

One can now use `taskset(1)` and limit the number of workers
accordingly.

With some uggly duplication because we just want a quick test.

Simplify logic.
Add comments to explain motives.

Taken and adapted from:
https://github.com/scikit-learn/scikit-learn/blob/e4bb9fa86b0df873ad750b6d59090843d9d23d50/sklearn/neighbors/_binary_tree.pxi#L513

Adapt Makefile for python3.9

Update dependencies

This makes use of a class of actors wrapping a numpy array of
the nearest neighbors indices.

Also:
 - Replace depth for leaf_size
 - Make the counter an argument
 - Correctly manage the scheduler lifecycle