[LTX-Video] fix attribute adjustment for ltx.

[sayakpaul]

What does this PR do?

I want to be able to do:

pipe = LTXPipeline.from_pretrained(..., transformer=None, vae=None)

with torch.no_grad():
    ... = pipe.encode_prompt(...)

And then use the prompt embedding outputs for downstream tasks.

Currently, LTX pipelines would fail to do this. This PR fixes that.

In a separate PR, I can look into refactoring the test-suite to test encode_prompt() in isolation as it's widely used for many trainers and other applications (cc: @hlky).

[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.