Fix datasets task 7309: pin pyarrow < 20.0 for geoparquet compatibility

[AlienKevin]

Summary

• The Dockerfile for datasets task 7309 pins pyarrow==20.0.0
• PyArrow 20.0 changed the default Arrow string type mapping, causing geoparquet geometry columns to report as large_string instead of string
• The pre-existing test test_parquet_read_geoparquet asserts dataset.features[feature].dtype == 'string', which fails with large_string
• This breaks all feature pairs for this task (both feature 1 and feature 2 tests fail on the same pre-existing test)

Fix

Change the pyarrow pin in the Dockerfile from:

RUN uv pip uninstall --system pyarrow && uv pip install --system "pyarrow==20.0.0"

to:

RUN uv pip uninstall --system pyarrow && uv pip install --system "pyarrow>=17.0.0,<20.0.0"

The Docker image (akhatua/cooperbench-huggingface-datasets:task7309) needs to be rebuilt and pushed to DockerHub after this change.

Test plan

• Rebuild Docker image with pinned pyarrow
• Oracle test passes for datasets task 7309 (both features)