perf: optimize extraction hot paths

[ryanfowler]

Summary

• Reuse epoch/root-scoped chrome aggregates across cleanup passes.
• Skip comment-specific scans when the fragment has no possible comment markers.
• Avoid repeated dynamic attribute/name work on ordinary DOM nodes.
• Const-specialize the common unfiltered scoring statistics traversal.
• Preserve custom related-heading recognition and add regression coverage.

Performance

Controlled cargo bench --bench pipeline -- 'extract/large/prose$' --noplot runs used the same worktree, lockfile, and benchmark inputs with 100 samples:

• main / clean HEAD: 45.314 ms median
• this branch: 40.537 ms median
• improvement: 10.53% lower wall time

Baseline and current runs were also profiled with perf record to identify cleanup, scoring, and attribute/name lookup hot paths.

Validation

• cargo fmt --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test (622 unit, 14 fixture, 130 Mozilla compatibility, 3 doctests)

Ready for review.