Handle scalar datasets too (scverse#2344)

* Handle scalar datasets too

After @ivirshup's pytables PR (scverse#2064) we started having issues with loading h5 files with scalar datasets, such as those created by CellBender (broadinstitute/CellBender#128). It is currently not an issue for the 10X h5 files for now since they don't have any scalars, however it'd be good to just handle scalars as well as arrays 1- to fix the cellbender file loading problem 2- to fix potential problems we might end up having if 10X h5 format includes scalar datasets.

* Add a scalar to the multiple_genomes.h5 test file

* Fixes scverse#2203

scanpy/readwrite.py

@@ -253,7 +253,7 @@ def _read_legacy_10x_h5(filename, *, genome=None, start=None):
 def _collect_datasets(dsets: dict, group: h5py.Group):
     for k, v in group.items():
         if isinstance(v, h5py.Dataset):
-            dsets[k] = v[:]
+            dsets[k] = v[()]
         else:
             _collect_datasets(dsets, v)