What does this PR do?

When --with_prior_preservation is enabled, the training batch concatenates
instance and class (prior) samples, so every per-sample tensor —
model_pred, target, sigmas, and therefore weighting — has shape
(2 * train_batch_size, ...).

Inside the loss block, model_pred and target are correctly split via
torch.chunk(..., 2, dim=0), but weighting was never chunked. This means:

• weighting (size 2B) is broadcast against model_pred_prior and
  target_prior (size B), producing a loss tensor of the wrong shape and
  applying incorrectly paired timestep weights to the prior loss term.
• The instance loss term also gets weights from the full unsplit weighting
  instead of only the instance-sample half.

The correct pattern already exists in train_dreambooth_lora_flux2.py:

weighting, weighting_prior = torch.chunk(weighting, 2, dim=0)

This PR applies the same fix to train_dreambooth_lora_flux.py and
train_dreambooth_lora_sd3.py, which were both missing it.

Fixes # (issue)

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