Support Numpy structured arrays containing an object

[ombschervister]

Hello!
There are problems when working with NumPy Structured arrays containing an object.

Given the NumPy composite type:
records_dtype = [('col1', object), ('col2', np.float64)]

In numpy, data is retrieved by the name of the column, but in zarr this does not work:

arr = np.array([('1', 1.0), ('22', 2.0), ('333', 3.0)], dtype=records_dtype)
print(arr['col1'])

1. When creating dataset an exception occurs:
   ds = zarr.array(arr)
   zarr.errors.MetadataError: error decoding metadata: Cannot change data-type for object array.

2. If you pass 'fill_value = None' to zarr.array(), the dataset is created and the values from it can be read:

   ds = zarr.array(arr, fill_value=None)
   print(ds[:])
   

   But data retrieval by column name does not work:
   print(ds['col1'])
   ValueError: cannot create an OBJECT array from memory buffer

[joshmoore]

Fixed by #813 (v2.9.4)