llama.cpp Adds TENSOR_SKIP to Bypass gate_up_exps

The llama.cpp repository has merged a pull request allowing models to skip gate_up_exps when TENSOR_SKIP is set. This optimization targets inference runtime efficiency for specific model architectures.