speedup hunyuan encoder causal mask generation

[dabeschte]

What does this PR do?

The original causal attention mask generation for the hunyuan encoder is very slow, especially when the tensor is created on the GPU, because it needs to make 1000s of calls (for loop over sequence length).

I tried to compile it...which works and makes it fast too, but compilation unfortunately also takes a long time when using a long sequence length.

This implementation is ~20-70x faster depending on the sequence lengths and since it is re-created for every SDPA, this accumulates to multiple seconds per step for larger videos

This PR is a duplicate of the PR to the original HunyuanVideo repository Tencent-Hunyuan/HunyuanVideo#208

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a GitHub issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• [ X] Did you write any new necessary tests?

Who can review?

not sure tbh

[a-r-r-o-w]

Hey, thank you so much! It was on my mind to address this after the integration as it looked weird, but I forgot about it :/ This is super cool

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[a-r-r-o-w]

@dabeschte Could you look into the failing test? We can merge after that :)

[dabeschte]

thanks for the quick review.
yeah, it really does look weird to loop over the sequence lengths - and even more so when you profile model :D

I had a small bug in how I used the testing framework (looking at you, Cursor ;))
I didn't want to install all dependencies to get the tests to run - but this time I tested it in isolation to ensure at least this test is working correctly.