Return empty attention weights when need_atten_weights = False (#91782)
# Summary
This PR updates the second return value from SDPA to return an empty tensor of size 0 not what it would be if need_attn_weights is True. Also updates the meta function to account for this change.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/91782
Approved by: https://github.com/cpuhrsch