Fix indexing and freshness in large repositories

[aglumova]

Summary

• Handle multibyte whitespace while screening secret-like Markdown assignments.
• Apply the same portable-path and binary-source rules to indexing, freshness checks, and Lens.
• Keep large repositories within explicit bounds of 100,000 candidates, 512 MiB of source, and a 30-second status scan.
• Point the published benchmark cases at the matching commits in the cleaned history.

Verification

• Rust regression tests for unsupported Git filenames and a NUL byte beyond the admission prefix passed.
• cargo fmt --check, cargo clippy --lib -- -D warnings, and the bundled benchmark corpus tests passed.
• A local repository with over 50,000 indexed files returned stale_count=0; its context preview returned code: ok and person: ok.

The project-document layer remains incomplete where source artifacts were deliberately skipped by the existing safety limits.