[Kernels]Updated Triton kernels into 2.1.0 and adding flash-decoding for llama token attention

[Kernels]Updated Triton kernels into 2.1.0 and adding flash-decoding for llama token attention #4368