Fix orc pytest failures

[galipremsagar]

Description

This PR fixes all 16 orc pytest failures.
Cpp bug:

// The conversion map tells the statistics system how to normalize
// column types before computing min/max. For ORC, all timestamps
// must be converted to nanoseconds because split_nanosecond_timestamp()
// in stats_enc.cu expects nanosecond input.

What happened without the fix: A datetime64[ms] column with min value 1970-07-11 07:00:35.528 (stored as 16527635528 ms since epoch) was passed directly to split_nanosecond_timestamp() which interpreted it as nanoseconds:


split_nanosecond_timestamp(16527635528)
  → treats as 16,527,635,528 nanoseconds
  → 16,527 milliseconds + 635,528 nanoseconds remainder
  → ORC stats report min = 16,527 ms = 16.527 seconds
  → 1970-01-01 00:00:16.527  ← WRONG (off by ~1,000,000x)
With the fix: The value is first converted to nanoseconds via time_point_cast:


16527635528 ms → 16,527,635,528,000,000 ns (via conversion map)
split_nanosecond_timestamp(16527635528000000)
  → 16,527,635,528 milliseconds + 0 nanoseconds remainder
  → ORC stats report min = 16,527,635,528 ms
  → 1970-07-11 07:00:35.528  ← CORRECT

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[mhaseeb123]

two-liner C++ changes look in order to me but requested a review from @vuule

[vuule]

Changes look good.
Is there an issue for this? would love to see the issue linked in the PR description and a short description of what the fix is.

[galipremsagar]

Changes look good. Is there an issue for this? would love to see the issue linked in the PR description and a short description of what the fix is.

Thanks @mhaseeb123 & @vuule ! This is a fix to pandas3 feature branch. I don't have a bug opened for this but updated this pr description to have the bug description. #21847 (comment)