Add a spatial index

[boydjohnson]

The key insight was that the nodes, ways, and relations could be sorted in the curve's order and then found via binary search at query time. The schema is modified for relations to include the minimum bounding box, because that is often difficult to find at query time.

Entity	Curve	Resolution	Domain	Indexed on	Notes
Nodes	ZCurve2D (Z-order / Morton)	1024 (curve default)	lon/lat, full WGS84 (-180..180, -90..90)	point (lon, lat)	Points have no extent, so a plain Z-curve suffices. Default WGS84 bounds.
Ways	XZ2SFC (extended Z-order)	SPATIAL_RESOLUTION = 18	XZ2SFC::wgs84(18)	minimum bounding box (min_x, min_y, max_x, max_y)	XZ2 handles entities with extent (boxes). Ordered by bbox index.
Relations	XZ2SFC (extended Z-order)	SPATIAL_RESOLUTION = 18	XZ2SFC::wgs84(18)	minimum bounding box (mbb [i32; 4])	Same curve as ways (way_curve()). Bbox stored in the archive.

osmflatc now uses RocksDB to do the ordering based on the curve's index. There are osmflatc arguments for configuring RocksDB.

--scratch-dir <SCRATCH_DIR>
Scratch RocksDB directory; use a fast SSD with ample space [default: output's parent]
--block-cache-mb <BLOCK_CACHE_MB>
Scratch RocksDB block cache (MiB); raise (e.g. 8192) for planet-scale [default: 512]
--write-buffer-mb <WRITE_BUFFER_MB>
Scratch RocksDB write buffer (MiB); raise for higher write throughput [default: 128]
--max-open-files <MAX_OPEN_FILES>
Max open SST files RocksDB may cache (-1 = unlimited). Must stay below the process open-file limit (raise it with ulimit -n); lower this if you hit "Too many open files" [default: -1]

Note: Some of the code included in this PR was written by Claude Code.

[boydjohnson]

This PR was talked about in #81

[boydjohnson]

I realized I messed up the --ids index. Before it was implicitly id ordered, as well as all the nodes, ways, and relations. So before Entity => Id was O(1), and then binary search for Id => Entity.

With this PR, Entity => Id is still O(1) but there is no binary search for Id => Entity.

I am going to add an index that will do the --reverse-ids.