Setting random_state and multi-threading

Hi,

I wanted to report a bug I’ve noticed in setting random_state. The documentation regarding UMAP reproducibility (https://umap-learn.readthedocs.io/en/latest/reproducibility.html) seems to be incorrect in saying that setting random_state will halt multi-threading. When running UMAP on the same data with the same random_state set, the results vary when I allocate a single core vs. many cores (16).

Parameters: UMAP(n_components=3, a=2.0, b=1.5, random_state=42)

First instance components with single core: (10.861118, 3.735592, 8.769404)
First instance components with 16 cores: (11.034830, 4.050596, 8.633882)

Info:
Python 3.9.7
umap-learn 0.5.2
numpy 1.21.2
numba 0.55.1

Thanks!

Start with the reproducibility documentation linked in the issue and run UMAP with n_components=3, a=2.0, b=1.5, and random_state=42 using one core and 16 cores. Compare the reported component values and determine whether the implementation or the documentation needs to change so the reproducibility claim matches observed behavior.