fix bug for partial_fit when the number of batch samples is less than n_comp

[DingWB]

• Closes #
• Tests included or not required because: n_components must be less or equal to the number of samples, otherwise it would throw an error, for example, ValueError: n_components=100 must be less or equal to the batch number of samples 40. This error usually happens on the last chunk of the partial_fit.

• Release notes not necessary because:

For example, my adata.shape[0] is 1041 and I run IncrementalPCA sc.tl.pca(adata, n_comps=100, chunked=True,chunk_size=1000), and I got an error: ValueError: n_components=100 must be less or equal to the batch number of samples 40 on scanpy/preprocessing/_pca.py:256