Improve meaningful content extraction accuracy

[ryanfowler]

Summary

• remove redundant document indexes, compact link rails, inline promotions, copy status text, and separate code line-number gutters with conservative multi-signal checks
• preserve authored callouts, link sections, numeric code, protected semantic content, accordion headings, wiki navigation, and access barriers
• exclude interactive copy controls from visible H1 metadata without changing authored punctuation or deep-markup behavior
• bound the new rare-candidate probes for untrusted HTML
• add an original compact news-rail snapshot and focused positive, negative, deep, and adversarial regressions

Live audit

I sampled the current Hacker News front page and representative article, blog, GitHub, Rustdoc, MDN, WordPress, and project-documentation pages. The changes generalize the repeated failure shapes instead of matching hostnames. Existing GitHub specialized extraction remained effective, so this change does not add a new site-specific parser.

Validation

• cargo fmt --check
• cargo fmt --manifest-path cli/Cargo.toml --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo clippy --manifest-path cli/Cargo.toml -- -D warnings
• cargo test (611 unit tests, 217 fixture tests, 14 public API tests, 130 Mozilla Readability tests, and 3 doctests)
• node tools/extractor-eval/index.mjs --all (134 fixtures: 0.95 recall, 0.97 noise rejection, 0.90 structural fidelity)
• cargo bench --bench smoke (no performance regression in all five workloads)

Review

Three independent subagent review rounds completed. All thirteen actionable findings were addressed with code changes and regression tests.