Minimal, reproducible code sample, a copy-pastable example if possible

import numpy as np
import zarr
a = np.ones((2,2)) 
z = zarr.open(zarr.NestedDirectoryStore("test.zarr"), 'w')
z['test'] = a

z = zarr.open(zarr.NestedDirectoryStore("test.zarr"), 'r')
print(np.array(z['test']))

arr = zarr.open(zarr.NestedDirectoryStore("path/to/existing/zarr/file.zarr"), 'r')
print(np.max(arr['volumes/raw']))

output with zarr 2.10.0:
array([[1., 1.],
[1., 1.]])
0

output with zarr 2.3.0:
array([[1., 1.],
[1., 1.]])
6733

Problem description

With the current version of zarr (2.10.0) pre-existing zarr files using NestedDirectoryStore are all zero when read.
With an older version (tested with 2.3.0) the correct values are loaded.
Creating a new zarr file using NestedDirectoryStore on 2.10.0 and reading it works correctly.

Version and installation information

Please provide the following:

• Value of zarr.__version__: '2.10.0'
• Value of numcodecs.__version__: '0.9.1'
• Version of Python interpreter: 3.8.11
• Operating system (Linux/Windows/Mac): Linux
• How Zarr was installed (e.g., "using pip into virtual environment", or "using conda"): pip install