[Feature] Support SDXL training

[okotaku]

Motivation

Support SDXL training.

Related PR: open-mmlab/mmengine#1361

Use cases (Optional)

mim train mmagic configs/stable_diffusion_xl/stable-diffusion_xl_pokemon_blip.py

Results

stable-diffusion_xl_pokemon_blip

weight: https://drive.google.com/file/d/1fzJwdW2yBjEqBDdQ1ILvQ23eVBK43jLO/view?usp=sharing

stable-diffusion_xl_lora_pokemon_blip

Checklist

Submitting this pull request means that,

Before PR:

• I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• New functionalities are covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with some of those projects.
• CLA has been signed and all committers have signed the CLA in this PR.

[okotaku]

Why do so many cpu tests fail, including some that have not been changed?

https://github.com/open-mmlab/mmagic/actions/runs/6258709202/job/16993265761?pr=2040#step:10:4026

[codecov]

Codecov Report

Attention: 39 lines in your changes are missing coverage. Please review.

Files	Coverage Δ
mmagic/datasets/__init__.py	100.00% <100.00%> (ø)
mmagic/datasets/transforms/__init__.py	100.00% <100.00%> (ø)
mmagic/engine/hooks/__init__.py	100.00% <100.00%> (ø)
mmagic/models/archs/lora.py	15.06% <100.00%> (ø)
...gic/models/data_preprocessors/data_preprocessor.py	97.41% <100.00%> (ø)
mmagic/engine/hooks/visualization_hook.py	95.53% <85.71%> (+1.31%)	⬆️
mmagic/datasets/transforms/loading.py	91.10% <78.12%> (-2.04%)	⬇️
mmagic/datasets/transforms/sdxl.py	93.06% <93.06%> (ø)
...magic/engine/hooks/lora_checkpoint_to_save_hook.py	56.25% <56.25%> (ø)
...editors/stable_diffusion_xl/stable_diffusion_xl.py	13.20% <0.00%> (-0.27%)	⬇️
... and 1 more

📢 Thoughts on this report? Let us know!.

[okotaku]

I am fixing the test error on mmengine PR.
open-mmlab/mmengine#1365