The zarr layout that explore() uses has grown piecemeal across remote_data.py, mesh_store.py, explore_zarr.py, and quickstart.py. Each file makes its own assumptions about group names, attribute conventions, and chunking. There's no single document describing what a valid rtxpy scene zarr actually contains.

This makes it awkward for anything outside the codebase to produce a scene file, and it means new data layers keep inventing their own storage conventions instead of following a common pattern.

A spec document should cover what explore() already consumes and define conventions for data it doesn't persist yet:

• Elevation DEM (CF encoding, LOD pyramid levels, chunking)
• Spatial reference / CRS metadata
• Mesh geometries (triangles, B-spline curves, spheres) spatially partitioned by chunk
• Raster overlay layers (slope, aspect, vegetation, etc.)
• Wind velocity grids
• Hydrological flow data (MFD directions, accumulation, stream order)
• Weather / cloud cover grids
• Camera state and rendering defaults
• Tour keyframes and observer positions

This would also let someone prepare a scene offline — fetch data, run analysis, bake meshes, save everything to one zarr — and then explore() it later without network access.

Deliverable: spec document in docs/ and any code changes to validate or produce conforming files.