[Medformatik]

Retire the experimental junction previews. Navigation keeps its maneuver instructions, route shields, exit labels and lane arrows in the green panel. The ordinary street-level imagery viewer and its providers remain available.

We explored additional destination boards, schematic junction drawings and lane guidance projected onto approach photos. Fixing image rendering and maneuver positions helped, but reliable guidance required more than a plausible projection: archived images could show a different carriageway or a separated ramp, and camera pose, road layout and lane correspondence were often uncertain. A calibrated example did not establish reliable general coverage. The extra displays could therefore contradict or confuse the useful maneuver guidance.

The decision is to retire the feature and delete its experimental stack. This PR deletes the lower panels, navigation-specific photo acquisition/preparation and overlay code, worker/build wiring, settings, translations, photo fixtures and evaluation tooling. Restoration would need a new evidence-based design.

The useful work remains:

• Ordered geometry positions shared by maneuver countdown, step advancement and the map turn arrow, while provider travel distances and ETA retain their units.
• Initially usable, preferred, rejected and unknown lane states internally, including correct provider mapping and navigation-session persistence. The phone UI highlights usable lanes uniformly with the existing arrow and background treatment; it adds no checkmarks or lane borders.
• OSM lane enrichment for the green arrows, with incoming-carriageway checks, explicit rejection/conflict handling, route-scoped state, reroute recomputation, retries and attribution. OSM-only unmarked lanes do not gain invented arrows.
• Sign-text normalization and exit-label cleanup, useful regression fixtures and synthetic public routing-canary endpoints.

The branch is squashed to one signed commit on current main. There is no migration or new toggle; the retired settings cease to have an effect.

Comparable renders of the actual navigation components use a controlled synthetic example, the application theme and bundled font. They demonstrate the UI removal; they are not a full map replay or a device driving test.

Before: main, 390 px light	After: this PR, 390 px light

Before: main, 320 px dark	After: this PR, 320 px dark

Validation: the full portable suite passed serially: 24,874 passed, 277 skipped (RAYON_NUM_THREADS=4 pnpm test --maxWorkers=1, under the shared test lock). The standard parallel run hit unrelated generator time limits; the isolated four-test generator suite passed. Lint, all 32 workspace type tasks, policy, OpenAPI and dead-code checks passed before rebasing.

Main advanced with Android Auto during that run. After rebasing, all 1,217 focused navigation/car tests passed, and affected core, web and mobile types and dead-code checks passed. The shared phone/car lane presenter retains the new main behavior and evidence-source checks. The final lane-display cleanup passed 284 focused navigation/car tests and affected core, web and mobile type checks.

The rendered examples have no horizontal overflow or browser page errors. Final self-review and an independent code review covered retirement completeness, retained navigation behavior and unintended changes; review regressions were reproduced and fixed before publication. Database-backed checks, native-device driving and deployment were not performed.

Release and upgrade impact

Product impact: breaking

The additional junction previews and their settings are removed. No operator action or data migration is required; the green maneuver guidance and ordinary imagery viewer remain available. Product release information is recorded in release-notes/retire-junction-previews.md. The navigation implementation lives in private workspaces. A patch Changeset covers the extension CLI because its conservative bundled-workspace dependency graph includes this code; its commands and shipped templates are unchanged.