[Alistair-Afton]

Summary

• MapCache built biomes[] as a fixed 3x3 neighborhood anchored at region_x/16 and indexed it directly with map_block.region_offset. region_offset is actually relative to each block's own region_pos, so on embarks that span multiple midmap cells, blocks outside the anchor cell resolved their biomes to the wrong world region.
• 3dveins then compared measured layer heights against the wrong geobiome spec and aborted with Layer height change ... instead of ... on such embarks.
• biomes[] now covers the union of neighborhoods around every midmap cell the map touches, with a biome_index map from world region coords to slots. Block::biomeIndexAt resolves the offset through block->region_pos, matching Maps::getBlockTileBiomeRgn, probe, and changelayer.
• Exported Maps::getBiomeRgnPos (previously an internal helper used by getBlockTileBiomeRgn/ReadGeology), now also guarding world_data and the index range.

Testing

• Reproduced the issue on a real crossing embark (embark rect spanning midmap cells (5,1) and (6,1)): 3dveins aborted with Layer height change in layer 5 at (144 0 112): 3 instead of 4 on the old build.
• Same embark on the new build: full statistics/generate/write cycle completes with no abort, and Maps::getTileBiomeRgn agrees with the corrected resolution.
• Single-midmap-cell fortress: 3dveins still completes normally.
• prospector (heavy MapCache consumer) smoke-tested on the crossing embark.

Fixes #5880