perf: avoid processing discarded rows for negative nth_value

[lyne7-sc]

Which issue does this PR close?

• Closes #.

Rationale for this change

For negative n, NTH_VALUE only needs to retain the last n rows. However, each update currently converts and appends every row in the input batch before discarding rows that cannot affect the result.

This adds unnecessary work for large batches, especially when n is small.

What changes are included in this PR?

• Retain only the required portion of the existing accumulator state.
• Append only the trailing input rows that can affect the result.
• Apply the optimization to both trivial and ordered accumulators.

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

Benchmark

group                       main                                     optimized
-----                       ----                                     ---------
nth_value/ordered/n=-1      5154.01     6.4±0.19ms        ? ?/sec    1.00  1233.7±449.90ns        ? ?/sec
nth_value/ordered/n=-10     819.35     6.5±0.25ms        ? ?/sec     1.00      7.9±0.14µs        ? ?/sec
nth_value/ordered/n=-100    87.44     6.4±0.18ms        ? ?/sec      1.00     73.7±1.80µs        ? ?/sec
nth_value/ordered/n=1       1.00   241.0±24.66ns        ? ?/sec      1.04   251.0±22.94ns        ? ?/sec
nth_value/ordered/n=10      1.13  1375.2±281.33ns        ? ?/sec     1.00  1219.0±51.07ns        ? ?/sec
nth_value/ordered/n=100     1.03      8.4±0.74µs        ? ?/sec      1.00      8.2±0.18µs        ? ?/sec
nth_value/trivial/n=-1      6821.05     3.3±0.24ms        ? ?/sec    1.00   485.5±20.29ns        ? ?/sec
nth_value/trivial/n=-10     822.65     3.4±0.33ms        ? ?/sec     1.00      4.1±0.10µs        ? ?/sec
nth_value/trivial/n=-100    82.35     3.3±0.26ms        ? ?/sec      1.00     40.2±4.02µs        ? ?/sec
nth_value/trivial/n=1       1.00     97.0±1.56ns        ? ?/sec      1.03   100.1±10.38ns        ? ?/sec
nth_value/trivial/n=10      1.01   599.9±38.67ns        ? ?/sec      1.00   591.2±22.46ns        ? ?/sec
nth_value/trivial/n=100     1.06      4.5±0.50µs        ? ?/sec      1.00      4.3±0.19µs        ? ?/sec

[codecov-commenter]

Codecov Report

❌ Patch coverage is 77.77778% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.24%. Comparing base (c429919) to head (0adf1bc).

Files with missing lines	Patch %	Lines
datafusion/functions-aggregate/src/nth_value.rs	77.77%	0 Missing and 4 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #24453      +/-   ##
==========================================
- Coverage   81.24%   81.24%   -0.01%     
==========================================
  Files        1113     1113              
  Lines      392744   392744              
  Branches   392744   392744              
==========================================
- Hits       319090   319087       -3     
+ Misses      54900    54897       -3     
- Partials    18754    18760       +6     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.