generate_prediction_data logic to choose tree_type does not respect algorithm chosen

[Dicksonchin93]

in https://github.com/scikit-learn-contrib/hdbscan/blob/27f279c1b33b2729f7248e42e10fca1b6d700ec2/hdbscan/hdbscan_.py#L1234-L1240 the logic to choose tree type to generate prediction data instance does not respect the algorithm parameter chosen, for 'prims_kdtree' and 'boruvka_kdtree' tree_type should be 'kdtree' and for 'prims_balltree' and 'boruvka_balltree' tree_type should be 'balltree'

I can make a PR to change this logic to respect the algorithm parameter chosen if that sounds good to the stakeholders here.