fix the bug of pca mean not saving

[Sichao25]

Fix the bug in the optimized sparse input PCA function that PCA mean matrix is not correctly saved.

[Xiaojieqiu]

@Sichao25 have you double checked the means / PCs are saved for all types of different pca approaches correctly?

[Sichao25]

PCs are saved in all methods. In Incremental/arpack/randomized PCA we call sklearn methods so the PCA means are automatically saved in the PCA object. If the sklearn TruncatedSVD is used, there won't be a PCA mean matrix (because TruncatedSVD will not center the input).

[Xiaojieqiu]

in that case, please set the PCA mean matrix to be all zero. And the pca means should be saved in the adata.uns because we will need to create the reverse projection to make the perturbation predictions