Contents of precomputed_knn tuple

Nobody has claimed this yet.

I'm trying to feed to UMAP nearest neighbors computed with another method than pynndescent, and so I need to put the data in the format of precomputed_knn that UMAP accept: the tuple: (knn_indices, knn_distances, NNDescentObject)
Since I had trouble making an NNDescent object without giving it data, I tried giving it a fake NNDescent object, that returns true to isinstance(knn[2], NNDescent):

```python
import pynndescent
import numpy as np
import umap
import umap.plot
from umap.umap_ import nearest_neighbors
import matplotlib.pyplot as plt

n = 200
N = 3*n
y = np.random.rand(n, 60)
X = np.concatenate((y+20, y, y-20))
synthetic_labels = np.repeat([1, 2, 3], repeats=n)

random_seed = 10

knn = nearest_neighbors(
X,
n_neighbors=30,
metric='euclidean',
metric_kwds={},
angular=False,
random_state=random_seed,
)

knn_umap = umap.UMAP(n_neighbors=30, precomputed_knn=knn, random_state=random_seed, force_approximation_algorithm=True)
knn_umap.fit(X)

class MyNNDescent(pynndescent.NNDescent):
def __init__(self):
return

knn2 = (knn[0],knn[1],MyNNDescent())
knn_umap2 = umap.UMAP(n_neighbors=30, precomputed_knn=knn2, random_state=random_seed, force_approximation_algorithm=True)
knn_umap2.fit(X)

print("\033[1m"+"Are the embeddings for knn_umap and knn_umap2 the same?\033[0m")
print((knn_umap.embedding_ == knn_umap2.embedding_).all())

fig, ax = plt.subplots(1, 2, figsize=(13,7))
umap.plot.points(knn_umap, labels=synthetic_labels, ax=ax[0], theme='green')
umap.plot.points(knn_umap2, labels=synthetic_labels, ax=ax[1], theme='green')
ax[0].set_title("Precomputed knn 1st run", size=16)
ax[1].set_title("Precomputed knn 2nd run", size=16)
plt.show()
```

The result is that the 2 embeddings are the same, so it seems that the NNDescent object is not used at all.
Is my assumption correct, or is the object used in some particular cases, which would explain why it was included in the tuple?
I tried digging into the code, but I can't find the place where this object is used...

Start by tracing the UMAP.UMAP precomputed_knn path and the nearest_neighbors entry point, then inspect where the tuple's NNDescent object is accessed. Done means documenting whether that object is used and identifying the cases, if any, in which it affects processing.

Written by the indexing model from the issue text.