feat: footprint to tiles (MAPCO-11173)

[almog8k]

Question	Answer
New feature	✔
Documentation	✔
Tests added	✔

Further information:
Adds a new geometry-only API that computes, for a footprint (Polygon / MultiPolygon / Feature, holes supported) and an inclusive zoom span, all intersecting tiles on the WGS84 geodetic 2:1 grid — emitted lazily as compressed ITileRanges.

How the algorithm works:
The tile pyramid is a quadtree: tile (z, x, y) contains exactly tiles (z+1, 2x..2x+1, 2y..2y+1), so classifying one coarse tile can settle its entire subtree. Starting from the grid's two zoom-0 root tiles, each visited tile is classified by filtering the footprint's edge list against its bounds (Liang-Barsky clipping):

• No edges cross, center inside → the boundary doesn't pass through the tile, and a tile is a connected region, so it's uniformly inside. Its whole subtree is emitted arithmetically — one range per zoom level, zero further geometry work. This is how a coarse tile stands in for millions of deep ones.
• No edges cross, center outside → uniformly outside; the whole subtree is pruned.
• Edges cross → the boundary passes through, so the tile intersects (a polygon includes its boundary). Emit it and recurse into its 4 children, passing only the surviving edges — the edge list shrinks as the descent converges onto the boundary, keeping deep nodes cheap.

Recursion therefore only follows the footprint's boundary: total geometry work is O(boundary tiles at maxZoom) instead of O(bbox area × zoom levels). Interior bulk and exterior bulk are both settled at the coarsest possible level, which is also why memory stays flat — interior subtrees never get expanded, they're expressed as ranges. Point-in-polygon uses the even-odd ray-casting rule, which makes holes and MultiPolygon parts work with no special casing.

[github-actions]

🎫 Related Jira Issue: MAPCO-11173