[PERF] Avoid materializing not-equal pairs for keep first or last

[tunglambk]

PR Description

• select the first or last unequal right row from at most two candidates instead of materializing every unequal pair
• preserve the existing materialized-path ordering and null semantics
• cover duplicates, sorted and unsorted inputs, single-value inputs, nullable integers, datetimes, floating NaN, all-null data, and empty inputs

ELI5

For any left value, the first different right value is either the first right value or, if those are equal, the first value different from it. The same rule works backwards for keep="last", so the join only needs two candidates rather than a nearly Cartesian list of pairs.

Performance

End-to-end timings for 2,000 left and 2,000 right rows (best of three runs):

Cardinality	Keep	Before	After	Speedup
2	first	47.252 ms	3.541 ms	13.3x
2	last	50.935 ms	3.529 ms	14.4x
2,000	first	87.863 ms	3.451 ms	25.5x
2,000	last	83.081 ms	3.441 ms	24.1x

Correctness

• a direct-scan oracle covers the edge cases listed above for both keep modes
• an explicit regression test preserves the legacy output order
• 2,000 randomized int64, nullable-integer, and floating-NaN trials matched the materializing implementation
• keep="all" remains on the existing code path

Validation

• focused conditional-join suite: 259 passed, 1 skipped
• full suite: 1,358 passed, 5 skipped, 58 xfailed, 7 xpassed; the single sandbox DNS failure passed when rerun with network access
• all relevant pre-commit hooks passed on the changed Python files

This PR resolves #1651.

PR Checklist

1. PR from a feature branch on a fork.
2. Add a line to CHANGELOG.md under the latest version header.

Relevant Reviewers

• @ericmjl

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.86%. Comparing base (901f4b3) to head (d64e55e).
⚠️ Report is 218 commits behind head on dev.

Additional details and impacted files

@@            Coverage Diff             @@
##              dev    #1681      +/-   ##
==========================================
- Coverage   87.56%   85.86%   -1.70%     
==========================================
  Files          95      125      +30     
  Lines        6819    10018    +3199     
==========================================
+ Hits         5971     8602    +2631     
- Misses        848     1416     +568     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[samukweku]

thanks @tunglambk

kindly update the docs as well, informing users about this approach for != join

[tunglambk]

Addressed the review in d64e55e: the changelog now credits @tunglambk, and the conditional_join documentation describes the optimized single-!= path for keep="first" and keep="last". I also rebased onto the current dev. Locally, the full conditional_join test file passes (296 passed, 1 skipped), and the documentation build completes successfully. The CI documentation job reaches the preview deployment step, then receives a 403 when github-actions[bot] pushes to the upstream gh-pages branch.

[samukweku]

Thanks @tunglambk

[samukweku]

@tunglambk kindly merge