[Backend Tester] Reduce log verbosity / spam

[GregoryComer]

When running tests, deprecation warnings for export_for_training and "internal consistency verification was requested but not available" create a large amount of spam in the CLI and drown out actual test info during the run. Neither warning is particularly useful. I've suppressed the the export_for_training warning in the backend tester only and switched the tester to not request internal consistency verification since it's not compiled into pybindings by default. This gives much cleaner CLI output.

Before:

test_add_dtype_float32_xnnpack_static_int8_per_channel (test_add.Add.test_add_dtype_float32_xnnpack_static_int8_per_channel) ... /home/gregory/src/executorch/src/executorch/backends/test/harness/stages/quantize.py:50: FutureWarning: `torch.export.export_for_training` is deprecated and will be removed in PyTorch 2.10. Please use `torch.export.export` instead, which is functionally equivalent.
  captured_graph = export_for_training(artifact, inputs, strict=True).module()
/home/gregory/miniconda3/envs/executorch/lib/python3.12/site-packages/torchao/quantization/pt2e/utils.py:818: FutureWarning: `torch.export.export_for_training` is deprecated and will be removed in PyTorch 2.10. Please use `torch.export.export` instead, which is functionally equivalent.
  aten_pattern = torch.export.export_for_training(
/home/gregory/miniconda3/envs/executorch/lib/python3.12/site-packages/torchao/quantization/pt2e/utils.py:818: FutureWarning: `torch.export.export_for_training` is deprecated and will be removed in PyTorch 2.10. Please use `torch.export.export` instead, which is functionally equivalent.
  aten_pattern = torch.export.export_for_training(
/home/gregory/miniconda3/envs/executorch/lib/python3.12/site-packages/torchao/quantization/pt2e/utils.py:818: FutureWarning: `torch.export.export_for_training` is deprecated and will be removed in PyTorch 2.10. Please use `torch.export.export` instead, which is functionally equivalent.
  aten_pattern = torch.export.export_for_training(
/home/gregory/miniconda3/envs/executorch/lib/python3.12/site-packages/torchao/quantization/pt2e/utils.py:818: FutureWarning: `torch.export.export_for_training` is deprecated and will be removed in PyTorch 2.10. Please use `torch.export.export` instead, which is functionally equivalent.
... (it logs like 20x per test)

After:

test_add_dtype_float32_xnnpack (test_add.Add.test_add_dtype_float32_xnnpack) ... ok
test_add_dtype_float32_xnnpack_static_int8_per_channel (test_add.Add.test_add_dtype_float32_xnnpack_static_int8_per_channel) ... ok
test_add_f32_alpha_xnnpack (test_add.Add.test_add_f32_alpha_xnnpack) ... ok
test_add_f32_alpha_xnnpack_static_int8_per_channel (test_add.Add.test_add_f32_alpha_xnnpack_static_int8_per_channel) ... ERROR
...