Dataset.to_zarr() method writes array data even when passed compute=False

What happened:
When writing out a Zarr dataset from an xr.Dataset, I pass the "compute=False" flag to tell xarray to write out the metadata but not the array data. When I inspect the Zarr dataset in the file system, I find that the array data has been written.

What you expected to happen:
I expected to find the metadata for my Zarr in the file system, but not the array data.

Minimal Complete Verifiable Example:

import xarray as xr, numpy as np, pandas as pd
ds = xr.Dataset({"foo": [10,20,30,40,50,60,70,80,90]})
ds.to_zarr("/tmp/bug.zarr", compute=False, consolidated=True)

bash$ cd /tmp/bug.zarr/foo
bash$ ls -la
total 20
drwxr-x--- 2 user group 4096 Mar 26 16:05 .
drwxr-x--- 6 user group 4096 Mar 26 16:05 ..
-rw-r----- 1 user group   88 Mar 26 16:05 0        # <---- I don't believe this should be here
-rw-r----- 1 user group  312 Mar 26 16:05 .zarray
-rw-r----- 1 user group   48 Mar 26 16:05 .zattrs

Anything else we need to know?:

Environment:

Start with the Dataset.to_zarr() entry point and reproduce the minimal example using compute=False and consolidated=True. Done means the Zarr store contains metadata but does not contain the array data chunk shown under foo/0.