[dynamo] avoid graph break on repeat_interleave.self_int (#99528)
Address convit_base failure: https://github.com/pytorch/torchdynamo/issues/1886 mentioned in https://github.com/pytorch/pytorch/issues/93777
Also for models like EleutherAI/gpt-j-6B.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99528
Approved by: https://github.com/ezyang
