Add runtime cluster/reducer registry

[patcon]

Until now, reducer and clusterer algorithm were added in the code of red-dwarf. Trying a new algorithm in the pipelines meant editing the code of red-dwarf. This was annoying for trying new algorithms.

Now, the builtin algorithms as well as any new ones are managed in their own runtime registries.

In other words, this is now possible: (same goes for clustering algos)

from reddwarf.utils.reducer.registry import register_reducer

@register_reducer('umap')
def make_umap(**kwargs):
    import umap

    # Set sensible defaults for UMAP
    defaults = {
        'n_components': 2,
        'n_neighbors': 15,
        'min_dist': 0.1,
        'metric': 'euclidean',
        'random_state': 42,
        'n_jobs': 1  # For reproducibility
    }

    # Override defaults with any provided kwargs
    defaults.update(kwargs)

    return umap.UMAP(**defaults)

from reddwarf.implementations.base import run_pipeline
from reddwarf.data_loader import Loader

result = run_pipeline(
        votes=votes,
        reducer="umap",                    # Use our registered UMAP reducer
        reducer_kwargs={
            'n_neighbors': 20,
        },
        clusterer="hdbscan",                # Default hdbscan clustering
    )

Runnable:
https://gist.github.com/patcon/33d849f528128fe21ee3fcedc0c2ac35#file-test-py-L6-L48