[DreamBooth SDXL LoRA] Apply SNR min-weighting to prior preservation loss when snr_gamma is set

[Dev-X25874]

What does this PR do?

In examples/dreambooth/train_dreambooth_lora_sdxl.py, when training with
both --snr_gamma and --with_prior_preservation, the Min-SNR loss
weighting (Section 3.4 of https://huggingface.co/papers/2303.09556) is
applied to the instance loss but silently skipped for the prior preservation
loss. The prior loss is computed before the snr_gamma branch runs, so
mse_loss_weights is never multiplied into it — the prior loss always
falls back to plain MSE regardless of --snr_gamma.

This PR fixes the snr_gamma branch to also apply mse_loss_weights to
prior_loss, mirroring the same pattern already used for the instance loss.
A new test covering the --snr_gamma + --with_prior_preservation combination
is also added to test_dreambooth_lora.py.

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?
• Did you write any new necessary tests?

Who can review?

@sayakpaul

[Dev-X25874]

Hey @sayakpaul — would love to get your eyes on this when you have a moment! 🙏

This fixes a silent bug where mse_loss_weights from --snr_gamma was never applied to the prior preservation loss in train_dreambooth_lora_sdxl.py — so prior loss always fell back to plain MSE even when SNR weighting was explicitly requested. The fix mirrors the exact same pattern already used for instance loss, and I've added a test covering the --snr_gamma + --with_prior_preservation combination.

Happy to adjust anything based on your feedback!