feat(RVV): Optimize MatrixAdd, MatrixSub, and MatrixMax with vector intrinsics

[ihb2032]

This PR introduces RISC-V Vector intrinsic optimizations for the following element-wise matrix operations:

• MatrixAdd
• MatrixSub
• MatrixMax

Performance Gains

The RVV implementation shows substantial speedups compared to the scalar version. Here are some highlights from the test results:

• MatrixAdd: Achieved up to 6.36x speedup.
• MatrixSub: Achieved up to 6.21x speedup.
• MatrixMax: Achieved up to 13.48x speed-up.

Testing Environment

The tests were conducted using the same environment as in PR #3779.

# test_matrix_sub
widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024
Scalar time : 0.0966 sec
RVV time : 0.0156 sec, Speedup: 6.21x
test widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024: PASSED

# test_matrix_max
widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024
Scalar time : 0.2053 sec
RVV time : 0.0152 sec, Speedup: 13.48x
test widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024: PASSED

# test_matrix_add
widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024
Scalar time : 0.0970 sec
RVV time : 0.0153 sec, Speedup: 6.36x
test widthC4=1024, cStride=4096, aStride=4096, bStride=4096, height=1024: PASSED