Fix TSNE perpelexity parameter for sklearn 1.2.1

[himaghna]

[JacksonBurns]

@himaghna I found out that the issue is (currently) only in the tests themselves, and was able to patch it over like this. If you could check in the clustering if it is possible that the default perplexity (30.0, as of sklearn 1) would ever be less than the number of samples and then write some defensive test for it, we should be covered.

[JacksonBurns]

slight correction, it is now len(self.test_smiles) instead of n_features