`allow_single_cluster=True` gets overwritten for `cluster_selection_method=leaf`

[divyegala]

If the cluster tree is empty and allow_single_cluster=True, root is chosen as a cluster here https://github.com/scikit-learn-contrib/hdbscan/blob/master/hdbscan/_hdbscan_tree.pyx#L780

Now, if cluster_selection_epsilon == 0.0, the selected clusters get overwritten to be the empty leaves here https://github.com/scikit-learn-contrib/hdbscan/blob/4052692af994610adc9f72486a47c905dd527c94/hdbscan/_hdbscan_tree.pyx#L785

Finally, the root ends up getting deselected here https://github.com/scikit-learn-contrib/hdbscan/blob/4052692af994610adc9f72486a47c905dd527c94/hdbscan/_hdbscan_tree.pyx#L788

Is this intended behavior, or should root continue to be a cluster in this case?

[lmcinnes]

I believe this is a bug, and not the intended behaviour.

[divyegala]

@lmcinnes what is the expected behavior? I'm happy to submit a patch/fix

[lmcinnes]

If allow_single_cluster is true then presuming then the root should not be removed as a candidate; if there is simply a single cluster that slowly decays (e.g. root. == leaf) then it should return the same result as if the "eom" method selected the root cluster. A patch would be most welcome! Thank you!