v5.2.0 — navigate like a screen-reader user

Blindfold, level two — the research pass.

Real screen-reader users navigate by HEADINGS and LANDMARKS first; Tab is the fallback. A journey that only Tab-walks never sees the two failures a real user hits on every page. Now the blindfold loop navigates like they do:

• 5 new friction patterns (15 total): journey_heading_nav_broken (2.4.6), journey_landmark_missing (2.4.1), journey_mode_confusion (4.1.2), journey_duplicate_targets (4.1.2), journey_dynamic_stale (4.1.3).
• The prompt now tests navigation itself: list headings and landmarks from every snapshot and jump by them — a broken hierarchy is a finding, not an inconvenience.
• Worked example, run for real: docs/journey-example.md — the author ran a blindfold journey against a failing fixture strictly by the rules. Result: blocked, 0/100, six findings across six criteria, gave_up protocol exercised in the wild.

Eight suites green; verified from a clean PyPI install.