Pin ray version to 2.23.0

[longshuicy]

Update:

• Ran into error: pyarrow.lib.ArrowCapacityError: array cannot contain more than 2147483646 bytes, have 4380140751, update the the create_geotiff_images_dataset to read individual files and do repartition

• Ran into

File "/home/cwang138/miniconda3/envs/maple_py310_ray/lib/python3.10/site-packages/osgeo/gdal.py", line 3600, in VSIFWriteL
    return _gdal.VSIFWriteL(*args)
RuntimeError: too large buffer (>2GB)

Fix by update the the gdal write file method to write in chunks

---

The latest Ray 2.41.0 introduces error: pyarrow.lib.ArrowInvalid: offset overflow while concatenating arrays, consider casting input from `binary` to `large_binary` first.

The workaround for now is pinning Ray to 2.32.0.

---

Some of the related notes:

• Suspect the stitching shapefile step has large binary. I’m guessing the stitched shapefile might have a very large binary field (> 2GB) that triggered the error.

data_per_image = inferenced_dataset.groupby(
        "image_name").map_groups(ray_tile_and_stitch_util.stitch_shapefile, concurrency=args.concurrency)

• Ray discussion: [Arrow] Adding ArrowTensorTypeV2 to support tensors larger than 2Gb ray-project/ray#47832

[tcnichol]

I have checked and with this version pinned everything runs. Doing some more review, will leave more comments. Just wanted to let you know that this now runs with no errors.