The return type of EXTRACT

I think the result type of u32 is not enough.
From the spark,
```
spark-sql> SELECT EXTRACT(YEAR FROM TIMESTAMP '-2000-12-16 12:21:13');
-2000
```
But this is not related to this issue, I will open a follow up pull request and discuss the returned type.

_Originally posted by @liukun4515 in https://github.com/apache/arrow-datafusion/pull/1991#r824743599_

Start by reproducing the EXTRACT(YEAR FROM TIMESTAMP '-2000-12-16 12:21:13') query and compare its returned type and value with the Spark SQL example in the issue. Locate the EXTRACT implementation and its existing tests; done means negative years are represented without the reported u32 limitation and the relevant test coverage passes.

Written by the indexing model from the issue text.