BUG readthedocs gives different results on data pruning example

[MatthewSZhang]

The examples of redundancy and data pruning on readthedocs show different results with my local machine (macos m1).
It may be caused by that cython floating is different on different machines.
To mitigate this problem, reduce tol to make FastCan more sensitive to the redundancy.

[MatthewSZhang]

The problem is mainly caused by the randomness in Kmeans.
The Kmeans results will be different even seed is set on different OS. See lmcinnes/umap#153
The problem can be mitigate by increase the random test times.