Use NVIDIA's RAPIDS cuML variant of HDBSCAN

[JohnTigue]

The first deploy target for the Codex is Colab. RAPIDS runs well on Colab with a GPU. So, claro, use RAPIDS implementations of HDBSCAN.

Will need to figure out best way to handle platform detection. Separate notebook variants for different deploy targets (JupyterLab, Colab, etc.) might be appropriate.

Relevant articles

• Oct 06, 2021 GPU-Accelerated Hierarchical DBSCAN with RAPIDS cuML – Let’s Get Back To The Future