compute flops for scaled_dot_product_flash_attention

https://github.com/pytorch/vision/actions/runs/5941974400/job/16117254380

Failures start with 9c4f7389d0db7cfe7e8591ea920459673344aaa8, which is the first commit that used yesterdays (20230822) PyTorch nightly.

   =========================== short test summary info ============================
  FAILED test/test_extended_models.py::test_schema_meta_validation[vit_b_16] - AssertionError: assert not [(ViT_B_16_Weights.IMAGENET1K_V1, '_ops'), (ViT_B_16_Weights.IMAGENET1K_SWAG_E2E_V1, '_ops')]
  FAILED test/test_extended_models.py::test_schema_meta_validation[vit_b_32] - AssertionError: assert not [(ViT_B_32_Weights.IMAGENET1K_V1, '_ops')]
  FAILED test/test_extended_models.py::test_schema_meta_validation[vit_h_14] - AssertionError: assert not [(ViT_H_14_Weights.IMAGENET1K_SWAG_E2E_V1, '_ops'), (ViT_H_14_Weights.IMAGENET1K_SWAG_LINEAR_V1, '_ops')]
  FAILED test/test_extended_models.py::test_schema_meta_validation[vit_l_16] - AssertionError: assert not [(ViT_L_16_Weights.IMAGENET1K_V1, '_ops'), (ViT_L_16_Weights.IMAGENET1K_SWAG_E2E_V1, '_ops')]
  FAILED test/test_extended_models.py::test_schema_meta_validation[vit_l_32] - AssertionError: assert not [(ViT_L_32_Weights.IMAGENET1K_V1, '_ops')]
  ====== 5 failed, 658 passed, 1 skipped, 430 warnings in 511.39s (0:08:31) ======

It is not obvious from the error message, but the failure here comes from the fact the number that we have on record, e.g.

https://github.com/pytorch/vision/blob/11e49de410ec84ec669293a91dfaa13a53c9bc47/torchvision/models/vision_transformer.py#L392

no longer matches what we calculate.

cc @seemethere