Fix Qwen Image DreamBooth prior-preservation batch ordering

[azolotenkov]

What does this PR do?

Fixes the prior-preservation batching issue in the Qwen Image DreamBooth LoRA script.

PR #13396 corrected the repeat count under --with_prior_preservation, but the script still repeats prompt embeddings with repeat(...), which interleaves the batch as [inst, class, inst, class, ...]. The dataloader and later torch.chunk(..., 2, dim=0) logic assume grouped ordering [inst1..instB, class1..classB].

In addition, weighting was not chunked alongside model_pred and target, which causes a batch mismatch for train_batch_size > 1.

This update:

• uses repeat_interleave(..., dim=0) for prompt_embeds and prompt_embeds_mask
• chunks weighting in the prior-preservation loss path

Reproduction:

accelerate launch \
  examples/dreambooth/train_dreambooth_lora_qwen_image.py \
  --pretrained_model_name_or_path hf-internal-testing/tiny-qwenimage-pipe \
  --instance_data_dir /dataset/instance_nocaptions \
  --instance_prompt "photo" \
  --with_prior_preservation \
  --class_data_dir /dataset/class \
  --class_prompt "class photo" \
  --num_class_images 2 \
  --resolution 64 \
  --train_batch_size 2 \
  --max_train_steps 1 \
  --learning_rate 5e-4 \
  --scale_lr \
  --output_dir /tmp/qwen-prior-b2-smoke

This fails with: RuntimeError: The size of tensor a (4) must match the size of tensor b (2)

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?

@sayakpaul