perf: keep raw dit image tensors on cpu until vae preprocess.

[chuanhaoyu]

Description

Keep raw DiT uint8 image tensors on CPU in DiTForwardInput::to(), and only move compute tensors to NPU.

VAEImageProcessor::resize() still uses a CPU Lanczos kernel, so moving images / images_list / mask_images / control_image / last_images to NPU early caused an extra NPU->CPU->NPU round-trip. This change avoids that copy for large and multi-image inputs such as Qwen Image Edit Plus.

Prompt embeddings, latents, masked_image_latents, and prompt_audio still follow the original .to(device) behavior.

Related Issues

N/A

Change Type

• Bug fix
• New feature
• Performance improvement
• Refactor
• Documentation
• Test
• Build or CI

Pull Request Checklist

Thank you for contributing to xLLM. Before requesting review, please make sure the following items are complete.

PR Title and Commit Messages

• The PR title and each commit message follow the xLLM commit format: <type>: <subject>.

Allowed types: feat, bugfix, docs, test, refactor, chore, style, revert, perf, model, build, release.
The subject should use clear English, start with a verb, include at least 4 words, and end with ..

Pre-commit Checks

• I have installed pre-commit by running pip install pre-commit or an equivalent command.
• I have installed the hooks with pre-commit install.
• I have run pre-commit run --all-files and fixed any reported issues.

If you are unsure how to set up pre-commit, see the pre-commit documentation.

Self Review

• I have self-reviewed the code according to .agents/skills/code-review/references/custom-code-style.md, especially code written or assisted by AI.
• I have rebased this PR onto the latest main branch.

Build and Test Coverage

• Tests have been added or updated as needed.
• CUDA: python setup.py build test has passed on a CUDA machine.
• NPU: python setup.py build test has passed on an NPU machine.
• MLU: python setup.py build test has passed on an MLU machine.

Reviewer Notes

This PR only changes xllm/core/runtime/dit_forward_params.h.

Validated on NPU with Qwen Image Edit Plus:

• single-image and multi-image requests succeed
• raw images_list stays on CPU before VAE preprocess
• condition_img / vae_img still land on NPU after preprocess
• preprocess latency does not regress

NPU Lanczos resize is out of scope. Flux Fill / Flux Control / Wan I2V share the same raw-image fields, but dedicated end-to-end smoke was not run because those model services were not available in this environment.