Multi-mst

Major API update to support multiple down-stream tasks. The classes now only construct the manifold models. Member functions are added to compute `umap` layouts or detect `hdbscan` clusters from the manifold. Especially the approximate nn-descent-based k-MST can speed up computing `hdbscan` clusters on high-dimensional data.

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.2.0

- Support numpy >= 2.0
- No longer force numba compile on import

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.1.4

- Added a nearest-neighbour descent based variant for the k-MST algorithm. This version creates approximate k-MSTs and is much faster on datasets that do not behave nicely with KD-Trees. The compute time and accuracy are evaluated in new benchmark notebooks.
- Keyword arguments for UMAP are no longer taken from `**kwargs`. Instead use the named `umap_kwargs=dict(...)` keyword argument.

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.1.3

- Update Github Actions to also publish pure python wheel.

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.1.2

- Updated Github Actions to create Releases automatically when version tags are pushed.

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.1.1

Initial Release.

**Full Changelog**: https://github.com/vda-lab/multi_mst/commits/v0.1.0