Update xee dataflow example

This updates the xee dataflow example to prevent users from accidentally deleting storage bucket when running the example.

This is a really simple fix for a bug in a recent [push to gcsfs](fsspec/gcsfs#608) paired with some [logic in the zarr library for writing datasets](https://github.com/zarr-developers/zarr-python/blob/df4c25f70c8a1e2b43214d7f26e80d34df502e7e/src/zarr/v2/storage.py#L567) which allows users to accidentally remove their bucket if writing to the root of a cloud storage bucket. This is problematic because users may have other data in a cloud storage bucket they may try to write to and accidental deletion of the bucket removes everything.

Changes in this PR include:
1. pinning the `gcsfs` version to `<=2024.4.0` before the PR that introduced the bug
2. point to write to subdirectory on the bucket in the example

PiperOrigin-RevId: 655683820

examples/dataflow/README.md

@@ -104,7 +104,7 @@ This example is focused on pulling data from Earth Engine, transforming the data
 ```shell
 python ee_to_zarr_dataflow.py \
   --input NASA/GPM_L3/IMERG_V06 \
-  --output gs://xee-out-${PROJECT} \
+  --output gs://xee-out-${PROJECT}/output/ \
   --target_chunks='time=6' \
   --runner DataflowRunner \
   --project $PROJECT \

pyproject.toml

@@ -44,7 +44,7 @@ tests = [
 dataflow = [
   "absl-py",
   "apache-beam[gcp]",
-  "gcsfs",
+  "gcsfs<=2024.4.0",
   "xarray-beam",
 ]
 examples = [