fix: keep gene order in batched highly_variable_genes without n_top_genes

[haeganm]

• Closes highly_variable_genes with batch_key and subset=True keeps the wrong genes when n_top_genes is not set #4370
• Tests included or not required because:
• Release notes not necessary because:

With batch_key set, the per-batch results are combined with df.groupby(df.index), which sorts the genes by name. #3042 restored the original order, but only in the n_top_genes branch. With the default mean/dispersion cutoffs the frame stayed name-sorted, so subset=True applied the mask by position and kept the wrong genes (on pbmc3k with two batches only 269 of the 1857 kept genes were actual HVGs). inplace=False also returned the rows in sorted order.

The fix reindexes to adata.var_names once after both branches instead of only inside the if.

New test test_subset_consistency_cutoffs runs seurat and cell_ranger with n_top_genes=None, with and without batch_key, on data with unsorted var_names. The two batched cases fail on main and pass here, the unbatched ones pass on both.