Handle NaN in Parquet statistics

[aalbu]

Fixes #4267

[aalbu]

Our ORC stats don't allow NaN. See https://github.com/prestosql/presto/blob/master/presto-orc/src/main/java/io/prestosql/orc/metadata/OrcMetadataReader.java#L337.

[findepi]

LGTM

do we have storage formats smoke tests to tests full write-read path?
this should be run for ORC as well, even though there isn't a problem to fix there

(like io.prestosql.plugin.hive.TestHiveIntegrationSmokeTest#testNaNPartition)

[aalbu]

do we have storage formats smoke tests to tests full write-read path?
this should be run for ORC as well, even though there isn't a problem to fix there

I did write a round-trip test, but I wasn't able to create stats with NaN, so it didn't fail without the fix 🤨 Inserting data through Presto didn't do it for me and I couldn't figure out how to insert NaN through Hive.

[findepi]

I did write a round-trip test, but I wasn't able to create stats with NaN, so it didn't fail without the fix

Since we use hive's writer, we depend in hive/parquet version we bundle.
Can you check whether the test would fail (before the change) with some old Presto version, like .172 (used by Athena)?

[aalbu]

Can you check whether the test would fail (before the change) with some old Presto version, like .172 (used by Athena)?

I have generated Parquet files with Athena and Spark, but I have not had any luck creating NaN file stats. I have added the test anyway.

[findepi]

Merged, thanks!