Optimization: Run a big MoE on a smaller GPU by keeping only the active units lit.

[andreampiovesana]

can we implement this optimization?
https://github.com/Luce-Org/lucebox-hub/tree/main/optimizations/spark

[luciopalmieri]

can we implement this optimization? https://github.com/Luce-Org/lucebox-hub/tree/main/optimizations/spark

It would be a very interesting and useful optimization.