[Fix][Dev] Typo fix for our workflow and enhance lop3 decode to support scaling

[LeiWang1999]

This pull request primarily focuses on enhancing the GPU intrinsic functions and updating the workflow configuration. The key changes include adding new decoding functions with scaling and offset capabilities, modifying the workflow configuration, and updating submodule references.

Enhancements to GPU Intrinsic Functions:

• Added new decoding functions with scaling and offset capabilities in bitblas/gpu/intrin/lop3.py. These functions include decode_i4_to_f16_scale_offset, decode_i4_to_f16_scale_zeros_original_offset, decode_i4_to_f16_scale_zeros_rescale_offset, and decode_i2_to_f16_scale_zeros_original_offset. [1] [2] [3] [4]
• Introduced get_func_arguments helper function to streamline the arguments passed to external functions.
• Updated the fast_decode_impl function to use the new helper function and added offset factors for buffers. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

Workflow Configuration:

• Changed depends-on to needs in the .github/workflows/benchmark.yml file to improve workflow dependencies.

Submodule Update:

• Updated the submodule reference for 3rdparty/tvm to a new commit.