Benchmark fresh and overlapping wheel installs

[zaniebot]

Wheel installation costs depend on the package layout, link mode, and whether files already exist. Exercise the production installer with published Flask, JupyterLab, NumPy, and SymPy wheels using both copies and hardlinks. Separate fresh installs from full-overlap replacements, and prepare distinct destination inodes outside the timed region so replacement cases reach the atomic merge paths. This covers astral-sh#19382, astral-sh#21468, and astral-sh#21478.

[codspeed]

Merging this PR will not alter performance

✅ 49 untouched benchmarks
🆕 16 new benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	WallTime	fresh_copy[flask]	N/A	7.5 ms	N/A
🆕	WallTime	fresh_copy[jupyterlab]	N/A	377.9 ms	N/A
🆕	WallTime	fresh_copy[numpy]	N/A	355.7 ms	N/A
🆕	WallTime	fresh_copy[sympy]	N/A	361.3 ms	N/A
🆕	WallTime	fresh_hardlink[flask]	N/A	2.1 ms	N/A
🆕	WallTime	fresh_hardlink[jupyterlab]	N/A	126.6 ms	N/A
🆕	WallTime	fresh_hardlink[numpy]	N/A	36.6 ms	N/A
🆕	WallTime	fresh_hardlink[sympy]	N/A	57.7 ms	N/A
🆕	WallTime	replace_copy[flask]	N/A	8.6 ms	N/A
🆕	WallTime	replace_copy[jupyterlab]	N/A	414.2 ms	N/A
🆕	WallTime	replace_copy[numpy]	N/A	399.9 ms	N/A
🆕	WallTime	replace_copy[sympy]	N/A	423.4 ms	N/A
🆕	WallTime	replace_hardlink[flask]	N/A	4.4 ms	N/A
🆕	WallTime	replace_hardlink[jupyterlab]	N/A	188.8 ms	N/A
🆕	WallTime	replace_hardlink[numpy]	N/A	124.5 ms	N/A
🆕	WallTime	replace_hardlink[sympy]	N/A	181.7 ms	N/A

---

_{Comparing zb/bench-real-wheel-install (f0b0e4a) with zb/bench-real-wheel-extraction (2398a89)¹}

Footnotes

1. No successful run was found on zb/bench-real-wheel-extraction (ca1d9e0) during the generation of this report, so c909300 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩