Parquet API: Could not convert timestamp before unix epoch to string/json

[AnthonyPoncet]

Describe the bug
Using the file::reader::SerializedFileReader in the parquet crate, I could not convert field of type timestamp with negative value (timestamp before UNIX epoch) to string. Issue exists with both micros and millis types. And I guess issue exist also while converting to JSON as it used same functions.

It seems to me that the i64 timestamp is converted into the Field enum as a u64. Then while using chrono to format it as string, it convert it back to i64 after divinding by 1000 where it should be before (see proposed fix bellow).

To Reproduce
In test test_convert_timestamp_to_string in file parquet/src/record/api.rs, change any tested date by something before 1970.
e.g: check_datetime_conversion(1969, 1, 2, 13, 12, 54);
-> Panic with the error No such local time

You could also create a parquet file with timestamp field and value before 1970 and then run

let file = File::open("<path>")?;
let reader = SerializedFileReader::new(file)?;
let mut row_iter = reader.get_row_iter(None)?;
while let Some(row) = row_iter.next() {
   let _ = row.get_column_iter()
                    .map(|c| c.1.to_string())
                    .collect::<Vec<String>>();
}

Expected behavior
Handling date before UNIX epoch but also do not panic.

Additional context
In the function convert_timestamp_millis_to_string changing let dt = Utc.timestamp((value / 1000) as i64, 0); to let dt = Utc.timestamp(value as i64 / 1000, 0); solve the issue.

Now I do not know the codeline enough to know if the fix should be the fix above or changing Field type to store a i64?

I could reproduce using both rust 1.63.0 and 1.64.0 on both Linux and Windows.

[AnthonyPoncet]

Hello @nevi-me
I am new to contributing to this repo, sorry if it is a dummy question. I opened a PR #2899, what are the next steps ?
Thanks

[alamb]

label_issue.py automatically added labels {'parquet'} from #2899