fix: don't drop multipolygon parts whose seed lands in an earlier part

[mailletf]

Problem

polygons_to_geohashes shares accepted_geohashes across the parts of a MultiPolygon, but the early continue skipped neighbour expansion, not just re-testing. A part whose seed cell had already been accepted while walking an earlier part terminated on its own seed and contributed nothing.

With A = rect(-73.60, 45.50, -73.55, 45.55) and B = rect(-73.590, 45.520, -73.520, 45.525) — B's centroid falls inside A — precision 7 returned A's 1369 cells where the union of the parts is 1479. 110 cells (7%) silently missing, and 66 of 1291 with fully_contained_only. No error raised.

Fix

Track a per-polygon visited set that gates the geometry test while accepted stays global, and always expand from any cell that intersects. Each part now gets a full walk regardless of what earlier parts covered, which also makes the separate rejected set redundant.

Two papercuts in the same loop:

• neighbor.to_string() cloned an already-owned String — 8 wasted allocations per visited cell.
• polygon.unsigned_area() was recomputed per candidate cell; it is O(vertices) and constant. Hoisting it is worth ~8% on verdun p9 with fully_contained_only.

Tests

Two regression tests, both failing on main: the union property, and order-independence of the parts.

[mailletf]

This change is part of the following stack:

• fix: don't drop multipolygon parts whose seed lands in an earlier part #30 ◀
  • perf: cover polygons by descending the geohash tree #31
    • perf: expand geohash sets on packed integers instead of strings #32
      • fix: cap the expansion hop count instead of hanging near the poles #33
        • fix: release the GIL while covering a polygon #34
          • chore: delete the dead handbrake cover #35
            • chore: cargo fmt #36
              • perf: add Arrow-native WKB and EWKB output #37
                • perf: add Arrow-native encode and coordinate decode #38
                  • perf: optionally dictionary-encode geog_id in the Arrow mapping output #39
                  • perf: cover large polygons across Rayon workers #40

_{Change managed by git-spice.}

[Copilot]

Pull request overview

This PR fixes an under-coverage bug in polygons_to_geohashes when processing multiple polygon parts (e.g., MultiPolygon parts) where a later part’s BFS would previously terminate early if its seed cell had already been accepted while walking an earlier part. It also removes some avoidable per-cell overhead and adds regression tests to prevent reintroducing the issue.

Changes:

• Introduces a per-polygon visited_geohashes set so each polygon part gets a full BFS walk regardless of globally accepted cells.
• Removes the per-polygon rejected_geohashes tracking and avoids unnecessary String allocations during neighbor expansion.
• Hoists polygon.unsigned_area() out of the inner loop and adds two regression tests (union property + order-independence).

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.