[training-scripts] Make more examples UV-compatible (follow up on #12000)

[linoytsaban]

make kontext and qwenimage also uv compatible, follow up on #12000

[linoytsaban]

@sayakpaul I wonder if qwen might need newer versions for transformers and/or peft, we don't have a specific requirements.txt for it

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[sayakpaul]

I wonder if qwen might need newer versions for transformers and/or peft, we don't have a specific requirements.txt for it

We can use the latest stable versions of both after running dummy runs? 👀

[linoytsaban]

btw @sayakpaul just realized from testing we need to add torchvision as well (to scripts from the original PR as well).
Also, to properly support features like loading a dataset from the hub, training with 8-bit Adam or training with prodigy, we need to add datasets, bitsandbytes, prodigyopt - though this goes somewhat against are policy of keeping the dependencies to the required minimum (and differs from the deps in requirements.txt). WDYT?

[linoytsaban]

failing test unrelated