Parquet: Incorrect min/max stats for int96 columns

[rahulketch]

Parquet write implementation currently emits incorrect min/max stats for int96 timestamp columns.

To Reproduce
Create a parquet file with int96 timestamp columns and observe the statistics.

Expected behavior

Additional context
See also:
apache/parquet-java#3242

[alamb]

Here is a related mailing list discussion on this topic: https://lists.apache.org/thread/6fm50b3pmh6mz659jb5wx5vzmvwccz1n

[alamb]

Here is my summary after the parquet sync call today:

• Pedantically, according to the current spec there is no defined ordering for Int96 types and thus arrow-rs can not be writing "incorrect" values (as there is no definition of correct)
• relevant portion of the spec:
  https://github.com/apache/parquet-format/blob/cf943c197f4fad826b14ba0c40eb0ffdab585285/src/main/thrift/parquet.thrift#L1079
• Practically speaking, arrow-rs is writing something different than Photon (Databricks proprietary spark engine)
• What Photon is doing arguably makes more sense (to use the ordering of the only logical type to use Int96)
• GH-7686: [Parquet] Fix int96 min/max stats #7687 brings arrow-rs into line with Photon

[alamb]

I also filed a PR in the parquet-format repo to try and clarify the meaning

• GH-502: Clarify Int96 status and add recommended ordering parquet-format#504

[alamb]

label_issue.py automatically added labels {'parquet'} from #7687

[alamb]

label_issue.py automatically added labels {'enhancement'} from #7687