There seems to be a regression in arrow-rs when decoding string columns that contain numbers. See https://github.com/apache/arrow-rs/issues/5095. This breaks all of the tests that use the Movies dataset, because some of the movie titles are numbers.
The arrow issue was fixed and should be released in arrow-rs 50. We'll have to wait until DataFusion updates to arrow 50+, which may be a couple of months.
WIP update to DataFusion 33.
Known Issues: