Resolve the WASM backend for the main-thread Points projection.
Points projection is bandwidth-bound and pairs with a zero-allocation
accumulator, so — unlike GSplats/Lines — it runs on the main thread
rather than a worker (offloading a memory-bound op would pay transfer
cost both ways for negligible compute savings). It still uses the WASM
kernels: data/points/projection.ts calls extract_3d_positions /
calculate_effective_radii on the module returned here, with the
uncapped TS reference selected for ndim > MAX_SUPPORTED_DIMS (the
same pickBackend routing the worker dispatchers use).
Resolve the WASM backend for the main-thread Points projection.
Points projection is bandwidth-bound and pairs with a zero-allocation accumulator, so — unlike GSplats/Lines — it runs on the main thread rather than a worker (offloading a memory-bound op would pay transfer cost both ways for negligible compute savings). It still uses the WASM kernels:
data/points/projection.tscallsextract_3d_positions/calculate_effective_radiion the module returned here, with the uncapped TS reference selected forndim > MAX_SUPPORTED_DIMS(the same pickBackend routing the worker dispatchers use).