[core] Track NaN counts in column statistics

[ArnavBalyan]

Purpose

• Spark, Flink and Iceberg all support IsNaN as a first class predicate and use NaN counts in file/partition pruning.
• Paimon's SimpleColStats only tracks min, max and null count today, there is no signal at the manifest layer to skip files.
• Add a nanCount field to SimpleColStats and update the collectors to count the nans, and further be used for engine level predicate pushdown.

Tests

• UT

[ArnavBalyan]

cc @JingsongLi could you PTAL thanks! :)

[JingsongLi]

This has a significant impact. I want to determine what changes have been made to the underlying storage, is it necessary?

[ArnavBalyan]

Will do small poc on the improvement of isnan pushdown and perf for stat collection, will revisit on mailing list for larger consensus