Is DensMAP parallelized? If we set random_state, are the results still stochastic?

This seems to be just as slow whether using `random_state=None` or a specific seed e.g. `random_state=42`.
```python
umap_trans = umap.UMAP(
densmap=True,
output_dens=True,
dens_lambda=self.dens_lambda,
n_neighbors=30,
min_dist=0,
n_components=2,
metric="precomputed",
random_state=random_state,
).fit(dm)
```
I'm also not sure if by setting `random_state=42` the results are actually deterministic.

Start by reproducing the supplied UMAP configuration with densmap=True, output_dens=True, metric="precomputed", and both random_state=None and random_state=42. Inspect the DensMAP implementation and its parallel execution path to determine whether seeded runs are deterministic and whether parallelization changes runtime. Done means documenting the observed behavior and the relevant conditions.

Written by the indexing model from the issue text.