llama.cpp Removes Optimized TG Path for MiniMax 01

A recent update to llama.cpp removes the optimized TG path in the MiniMax 01 implementation. This change is a technical indicator of the model's integration into the inference runtime.