Fix linear and matmul accuracy tests

[factnn]

Summary

• add a direct aten::linear accuracy/benchmark test with and without bias
• add a direct aten::matmul accuracy/benchmark test
• keep the existing baddbmm forward/backward tests, but remove their linear and matmul labels so they are not selected for those operators

Validation

• python -m py_compile src/kernelgenbench/accuracy/test_ops_with_benchmark.py
• git diff --check flagos/main...HEAD