[refactor] Remove additional Flux code

[a-r-r-o-w]

Some minor refactors to make implementations consistent.

• We already check for F.scaled_dot_product_attention in the attention processor init method (see here)
• Removes docstrings from intermediate blocks since they are explained in the main transformer implementation
• Additionally should we consider this comment and remove the NPU related logic in FluxSingleTransformerBlock? It might be backwards breaking though

[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.

[yiyixuxu]

Additionally should we consider #9751 (comment) comment and remove the NPU related logic in FluxSingleTransformerBlock? It might be backwards breaking though

let's remove it in this PR!
we could do deprecate if needed, but only for short-time, e.g like remove in 0.34.

[yiyixuxu]

other changes looks good to me!

[a-r-r-o-w]

Thanks @yiyixuxu, I've added a deprecation warning for v0.34.0

[yiyixuxu]

perfect!