[BUGFix] Introduce our own asser_close to allow few mismatch elements for some case

[LeiWang1999]

During ci testing with with_scaling=False, some test case may fail due to the range of uint4/int4 is too wide. This can lead to some mismatch elements when we leverage torch.testing.assert_close, even though the numbers are correct and make sense.

  File "/root/BitBLAS/testing/python/operators/test_general_matmul_bf16.py", line 193, in <module>
    matmul_torch_forward_weight_dequantize(1024, 1024, 1024, "bfloat16", "uint4", "float32",
  File "/root/BitBLAS/testing/python/operators/test_general_matmul_bf16.py", line 176, in matmul_torch_forward_weight_dequantize
    torch.testing.assert_close(permuted_inputs[-1], ref_result, rtol=1e2, atol=1e0)
  File "/opt/conda/lib/python3.10/site-packages/torch/testing/_comparison.py", line 1520, in assert_close
    raise error_metas[0].to_error(msg)
AssertionError: Tensor-likes are not close!

Mismatched elements: 2 / 1048576 (0.0%)

To allow few mismatch elements in some case, this pull request introduce our own assert_close function, which allow few tolerance for few mismatched arguments.

if not with_scaling:
    # when scaling is not enabled, we should have some mismatch due to the scaling factor
    bitblas.testing.torch_assert_close(permuted_inputs[-1], ref_result, rtol=1e2, atol=1e0)