Some robustifying fixes on Atol fit and tests for vectorizers

[martinroyer]

This introduces fixes that allow Atol to not crash on fitting, essentially by adding random points to the target. This is a follow-up on PR #1017 that can now be dropped.

Option: I add a series of test that only test Atol for now, but are ready to be extended to all vectorization methods. So one could work on another vector method and directly use the same test battery to see if there's a problem with this method's: - fit, - fit on empty diagrams, - transform, - transform on empty diagrams, - using sklearn set_output, - using sklearn compose with ColumnTransformer (not sure how this could fail with all the other tests but who knows). This test list comes as a sort of probe for a shared common interface of vectorizers, it could be enriched/edited on the way.

(naturally we could also work on the interface as a separate PR where we deal with the transform_one question too, see https://github.com/GUDHI/gudhi-devel/pull/1056#discussion_r1625083965.)