Parquet: AsyncArrowWriter to a file corrupts the footer for large columns

[DDtKey]

Describe the bug
AsyncArrowWriter to fs will lead to footer corruption if there is large column. However it works if we would use sync version of writer
See MRE below with a comment

To Reproduce

    #[tokio::test]
    async fn test_async_writer_to_file() {
        let col = Arc::new(Int64Array::from_iter_values([1, 2, 3])) as ArrayRef;
        // this column with random large strings will cause corruption of footer
        let col2 = Arc::new(StringArray::from(vec![generate_random_string(500000), generate_random_string(500000), generate_random_string(500000)])) as ArrayRef;
        // but this will work (column size is smaller)
        // let col2 = Arc::new(StringArray::from(vec![generate_random_string(50000), generate_random_string(50000), generate_random_string(50000)])) as ArrayRef;
        let to_write = RecordBatch::try_from_iter([("col", col), ("col2", col2)]).unwrap();

        let file = tokio::fs::File::create("/path/to/file/test.parquet").await.unwrap();
        let mut writer =
            AsyncArrowWriter::try_new(file, to_write.schema(), 0, None).unwrap();
        writer.write(&to_write).await.unwrap();
        writer.close().await.unwrap();

        let file = std::fs::File::open("/path/to/file/test.parquet").unwrap();
        let mut reader = ParquetRecordBatchReaderBuilder::try_new(file)
            .unwrap()
            .build()
            .unwrap();
        let read = reader.next().unwrap().unwrap();

        assert_eq!(to_write, read);
    }
    
    fn generate_random_string(length: usize) -> String {
        thread_rng()
            .sample_iter(&Alphanumeric)
            .take(length)
            .map(char::from)
            .collect()
    }

Reader fails with:

called `Result::unwrap()` on an `Err` value: General("Invalid Parquet file. Corrupt footer")

That's definitely something with writer, because I've tested other tools (e.g parquet-fromcsv) and the same files were written well. In addition, looks like sync writer also works fine.

Expected behavior
The output file shouldn't be corrupted, like with sync writer

Additional context

[tustvold]

label_issue.py automatically added labels {'parquet'} from #4527