fix: Fix incorrect parquet statistics written for UInt64 values > Int64::MAX

[nameexhaustion]

Fixes #16683, #15323

[nameexhaustion]

Test failure due to pip

[codspeed-hq]

CodSpeed Performance Report

Merging #16766 will improve performances by 17.4%

_{Comparing nameexhaustion:pq-statistics (7efd2fd) with main (2398b47)}

Summary

⚡ 1 improvements
✅ 36 untouched benchmarks

Benchmarks breakdown

	Benchmark	main	nameexhaustion:pq-statistics	Change
⚡	test_groupby_h2oai_q1	2.7 ms	2.3 ms	+17.4%

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.31%. Comparing base (a7f9c8d) to head (7efd2fd).
Report is 5 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16766      +/-   ##
==========================================
- Coverage   81.32%   81.31%   -0.01%     
==========================================
  Files        1423     1424       +1     
  Lines      187177   187209      +32     
  Branches     2721     2726       +5     
==========================================
+ Hits       152214   152228      +14     
- Misses      34468    34484      +16     
- Partials      495      497       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[deanm0000]

Not trying to sound like I'm criticizing but this doesn't really fix the issue even as it does avoid them in the future. The fix, I think, would be to add a check in the reader that if min_value is greater than max_value then ignore the stats and possibly issue a warning that the stats are invalid.

[deanm0000]

This is what I had in mind from my last comment. #16776

[nameexhaustion]

That's a good point, I imagine there are parquet files in the wild that have incorrect statistics (perhaps by us 😝), that some basic validation could help with.