Fix missing latents_bn_std dtype cast in VAE normalization

[adi776borate]

What does this PR do?

Follow up to #13050

This PR addresses an unintentional dtype/device mismatch during the VAE image encoding step in Flux2 pipelines.

Currently, inside _encode_vae_image, latents_bn_mean is explicitly cast to match image_latents, but latents_bn_std is left in its default buffer state.

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.
• Did you write any new necessary tests?

Who can review?

@asomoza
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[yiyixuxu]

thanks!

[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.

[adi776borate]

Thanks for the review! It should be ready to merge whenever convenient.

[iwr-redmond]

@asomoza @yiyixuxu what else needs to be done prior to merging this PR? It is a blocker for #13050.

[asomoza]

@iwr-redmond we merged the other PR so its not a blocker anymore, now this one needs to be updated

[dg845]

Thanks!