Stxtics03/vrgrid-26 ? reverse-engineered prompt

Build me a Python project for adaptive 2.5D LiDAR mapping called VRgrid.

I want it to take LiDAR frames and semantic labels, project them into a fixed memory map, and use finer resolution near the sensor and coarser resolution farther away. It should keep uncertainty honest when cells get merged or split, remove stale traces from moving objects, and let me query the map by world coordinates without caring about the grid size.

Please make the whole thing deterministic, CPU based, and easy to run on a normal machine. Include a small demo or pipeline for processing a sequence, plus tests that check the map stays consistent, merges and splits behave correctly, and the output is stable across runs. If you need to look up current docs online for any libraries or best practices, go ahead.