[release/2.6] unittest fixes

[dnikolaev-amd]

Fixes SWDEV-546357

torch.cuda.is_tf32_supported Is NOT public - Fixed by adding is_tf32_supported to __all__[]

test_public_bindings.py::TestPublicBindings::test_correct_module_names

AssertionError: Tensor-likes are not close! - Fixed by expecting AssertionError if ROCm

test_matmul_cuda.py::TestFP8MatmulCudaCUDA::test_float8_basics_cuda

RuntimeError: "index_cuda" not implemented for 'Float8_e8m0fnu' - fixed by converting to float16 for str(tensor) (like other fp8's)

test_quantization.py::TestFloat8DtypeCUDA::test_to_string_cuda_float8_e8m0fnu

OverflowError: Python integer 128 out of bounds for int8 - Fixed by skipping if numpy >=2.1

test_quantization.py::TestDynamicQuantizedOps::test_qlinear
test_quantization.py::TestDynamicQuantizedOps::test_qlinear_legacy
test_quantization.py::TestQuantizedLinear::test_qlinear
test_quantization.py::TestQuantizedLinear::test_qlinear_leaky_relu
test_quantization.py::TestQuantizedLinear::test_qlinear_relu
test_quantization.py::TestQuantizedLinear::test_qlinear_tanh
test_quantization.py::TestQuantizedLinear::test_qlinear_with_input_q_dq_qweight_dq_output_fp32

AssertionError: Tensor-likes are not equal! - Fixed by tolerance adjusting for MI350

test_scatter_gather_ops.py::TestScatterGatherCUDA::test_scatter_reduce_sum_cuda_complex128
test_scatter_gather_ops.py::TestScatterGatherCUDA::test_scatter_reduce_sum_cuda_complex64
test_scatter_gather_ops.py::TestScatterGatherCUDA::test_scatter_reduce_sum_cuda_float32
test_scatter_gather_ops.py::TestScatterGatherCUDA::test_scatter_reduce_sum_cuda_float64

ValueError: grad_value Test error is greater than threshold - fixed by adjusting fudge_factors['grad_value'] for MI350

test_transformers.py::TestSDPACudaOnlyCUDA::test_mem_efficient_attention_attn_mask_vs_math_ref_grads_batch_size_1_seq_len_q_2048_seq_len_k_8_head_dim_8_is_causal_False_dropout_p_0_0_float32_scale_l1_cuda_float32
test_transformers.py::TestSDPACudaOnlyCUDA::test_mem_efficient_attention_vs_math_ref_grads_batch_size_8_seq_len_q_2048_seq_len_k_8_head_dim_128_is_causal_True_dropout_p_0_0_float32_scale0_cuda_float32

RuntimeError: Cannot sort dimension of length 8192 - fixed and renamed to test_sort_large_cuda_float16

test_sort_and_select.py::TestSortAndSelectCUDA::test_sort_large_cuda_uint8

[rocm-repo-management-api]

Jenkins build for b360c6e5fc98b356d1bbd90ac0c9ca3957bcf9cf commit finished as FAILURE
Links: Blue Ocean view / Build artifacts

[rocm-repo-management-api]

Jenkins build for eebde3bd1406a0c90462af66ef996913205c7fab commit finished as FAILURE
Links: Blue Ocean view / Build artifacts

[rocm-repo-management-api]

Jenkins build for e299809adbee1eb97154f1bd942e7b383d6ac7e4 commit finished as NOT_BUILT
Links: Blue Ocean view / Build artifacts

[rocm-repo-management-api]

Jenkins build for e299809adbee1eb97154f1bd942e7b383d6ac7e4 commit finished as FAILURE
Links: Blue Ocean view / Build artifacts

[jithunnair-amd]

LGTM, but please address upstreaming comments and take appropriate action as needed.