fix(benchmark): verify pyodide target

[mstykow]

Summary

• Verify pyodide/pyodide@86e27b0 with compare-outputs on the latest main-based branch and add a durable benchmark row plus regenerated docs/scan-duration-vs-files.svg.
• Fix the real regressions the target exposed: standalone wheel archives now assemble into top-level packages, assignment-style author = / copyright = metadata no longer bleeds into copyright or holder output, and placeholder-host URLs like https://.../script.py are filtered structurally.
• Keep the final reviewed compare artifact at .provenant/compare-runs/20260505T205118Z-pyodide-13754 for auditability; the verified end state is 53 vs 52 packages, 796 vs 718 dependencies, and 10.77s vs 135.33s wall-clock time.

Scope and exclusions

• Included: assembly/refiner/finder code fixes, focused regression tests, the new docs/BENCHMARKS.md row for pyodide/pyodide, and the regenerated benchmark chart/headline stats.
• Explicit exclusions: no golden expected fixtures changed.

Intentional differences from Python

• Keep extension-qualified wheel PURLs for committed .whl artifacts instead of collapsing them to less specific unqualified PyPI identities.
• Keep the fuller docs/conf.py documentation notice and holder (Pyodide contributors and Mozilla) rather than ScanCode's shorter Pyodide contributors rendering.
• Keep extra dependency visibility from .gitmodules and environment.yml, plus patch-header author recovery on committed patch files.