fix(nearby): fan out co-located map pins so each stop is visible and tappable

[ai-tiro]

What

Co-located map pins are now fanned apart so each stop is visible and individually tappable. Different modes serving one physical station share identical PTV coordinates — the Richmond train stop (route_type 0) and the V/Line "Richmond Railway Station" (route_type 3) are both stop_id 1162 at -37.82407, 144.99016. Rendered as-is, their circles stacked on a single pixel: only the top one was visible, and the tap hit-test (nearest stop by true coordinate) could never resolve to the other.

How

New pure helper spreadColocatedStops() (in ColocationSpread.kt):

• Buckets stops by coordinate (~1 m grid, so exact + near-exact duplicates group).
• Fans any shared-point group evenly around a small ~20 m circle, slot order sorted by route type then id → each stop's nudged position is deterministic and doesn't jitter between renders.
• Lone stops keep their exact coordinate; grouping runs on the already-filtered set so only genuinely overlapping pins are nudged.

MapLibreOpenPtvMap uses it as the single source of display positions for both rendering and the tap hit-test, so a fanned-out dot is selectable exactly where it's drawn (this was the subtle part — without updating the hit-test, the dots would look separated but both taps would still resolve to the same stop, since they share a true coordinate).

Why pixel-offset, not clustering

You picked the offset approach. Re-enabling MapLibre clustering would reintroduce the #124 bug it was deliberately removed to fix (supercluster dropped stops at zoom extremes — metro vanished zoomed out, empty tiles zoomed in). The offset is a few metres, well under one stop's spacing, and keeps every stop visible at every zoom.

Testing

• Unit tests (ColocationSpreadTest, 4 cases): lone stop unchanged; two co-located stops fanned to distinct points each ~20 m from the shared point and ~40 m apart; fan-out deterministic and independent of input order; genuinely distinct nearby stops not merged.
• Manual QA on the AOSP emulator at Richmond (geo-fixed), Train + V/Line filters on: the V/Line pin renders as a distinct dot offset from the shared station point (the train pin sits at the true coordinate under the user-location dot), and tapping the offset V/Line pin opens the V/Line stop sheet — confirming it's independently selectable. Screenshots attached.
• Full :feature:nearby unit suite green; :app:assembleDebug green.

Discovered / notes

• The map pin list is not deduped (the ViewModel emits both stops), so the fan-out engages as intended.
• Unrelated pre-existing quirk noticed while testing: the Nearby tap-sheet for this stop lists the train serving routes (Alamein/Belgrave/…) alongside V/Line departures. Not touched here — flagging in case it's worth its own issue.
• The ~20 m separation is intentionally subtle; it only becomes visually obvious at street-level zoom. The emulator's scriptable zoom is coarse, so the attached "after" shot is at high zoom.

Fixes #172

[ai-tiro]

_{Uploaded via pr-attach.}

[github-actions]

Debug APK: app-debug-88e4d74cc42e9ed446336779f62f7556e63f0b6b.apk (built from 88e4d74 at 2026-06-20 06:07 UTC)

_{Requires GitHub login. Artifact expires after 3 days.}

[ai-tiro]

Superseded by #179, which combined these commits and has merged to master. Closing as superseded.