[BUG] AgglomerativeClustering: Duplicate data samples cause incorrect cluster assignments

[rsshah1993]

Describe the bug
Getting unexpected clustering results from AgglomerativeClustering.

Steps/Code to reproduce bug

import numpy as np
from cuml import AgglomerativeClustering

features = np.array([[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [2.0, 2.0, 2.0]])
AgglomerativeClustering(n_clusters=2).fit_predict(features)
>>> array([0, 1, 0], dtype=int32)

Expected behavior
Expecting outputs to be [0, 0, 1] or [1, 1, 0].

Environment details (please complete the following information):

• Environment location: Docker
• Linux Distro/Architecture: Ubuntu 18.04
• GPU Model/Driver: V100/455.32.00
• CUDA: 11.1
• Method of cuDF & cuML install: conda install -c rapidsai-nightly -c nvidia -c conda-forge -c defaults rapids=0.19 python=3.7 cudatoolkit=10.2 as part of a docker build.

[cjnolet]

I believe I know the cause of this bug. We assume absolute distance of 0 is on the diagonal in the pairwise distance matrix and so we set it to the max for the MST to converge to the correct solution. A reasonable fix for this case would be to do this only for the diagonal elements in order to support duplicate data samples.

Here's a small example of getting the correct solution by making the first two data samples slightly different.

>>> features = np.array([[0.0, 0.0, 0.001], [0.0, 0.0, 0.002], [2.0, 2.0, 2.0]])
>>> AgglomerativeClustering(n_clusters=2).fit_predict(features)
Label prop iterations: 3
Iterations: 1
2068,40,24,6,66,134
n_edges: 2
Finished dendrogram
array([0, 0, 1], dtype=int32)