[PERF] Speed up balanced NumPy pivot_longer reshapes

[samukweku]

PR Description

• Add a narrow fast path for balanced multi-value pivot_longer specifications backed by one homogeneous NumPy block.
• Slice already-contiguous value groups positionally instead of repeatedly selecting DataFrames and rescanning their dtypes.
• Preserve the existing path for extension arrays, mixed or fragmented blocks, and imbalanced specifications.
• Add exact ordering and dtype coverage for integers, floats, booleans, datetimes, and timedeltas in both appearance-order modes.

Representative median results on pandas 3.0.5:

Input	Before	After	Change
2,500,000 rows x 20 columns	79.7 ms	68.3 ms	14% faster
10,000 rows x 1,000 columns	20.5 ms	18.1 ms	12% faster
1 row x 100,000 columns	69.4 ms	31.4 ms	2.2x faster
1 row x 100,000 columns, 1,000 value groups	252.8 ms	33.4 ms	7.6x faster

Peak Python allocation was unchanged. Nullable, mixed, and imbalanced fallback cases remained within approximately 1% of the baseline.

Correctness was compared against dev over 720 generated combinations spanning empty, tiny, tall, and wide frames; balanced, interleaved, and imbalanced specifications; both ordering modes; dropna and ignore_index; and NumPy, object, categorical, nullable, string, and timezone-aware dtypes. All results matched element-for-element, including indexes and dtypes.

This PR contributes to #1656 without closing the profiling umbrella issue.

PR Checklist

1. Changes are on the focused issue-1656-stack-fast-paths branch.
2. Added an entry under CHANGELOG.md's Unreleased section.

Verification

• pixi run pytest tests/functions/test_pivot_longer.py: 103 passed, 6 xfailed, 1 xpassed
• Full suite: 1,351 passed, 5 skipped, 48 xfailed, 17 xpassed
• Focused Ruff check and format check passed
• git diff --check passed

Relevant Reviewers

• @ericmjl

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://pyjanitor-devs.github.io/pyjanitor/pr-preview/pr-1668/
Built to branch gh-pages at 2026-08-21 10:39 UTC. Preview will be ready when the GitHub Pages deployment is complete.