`torch.func.jvp` does not support flash_attention calculation

[pnotp]

🚀 The feature, motivation and pitch

When calling torch.func.jvp , the error is as follows:
[rank0]: NotImplementedError: Trying to use forward AD with _scaled_dot_product_flash_attention that does not support it because it has not been implemented yet.

Alternatives

No response

Additional context

No response

[jcaip]

Dupe of #165517