docs(int4): replace legacy llama mirror in examples

[glennko]

Summary:

• Replaces the legacy aleksickx/llama-7b-hf example checkpoint with Qwen/Qwen2.5-0.5B in the INT4 notebook and advanced any-model docs.
• Adds compatibility guidance that legacy mirrored checkpoints can fail on newer transformers versions.
• Updates the INT4 README note to direct users to maintained checkpoints.

Validation:

• python -m json.tool examples/features/int4_finetuning/LLaMA_lora_int4.ipynb >/dev/null
• docs build was not run in this branch because docs dependencies are not installed in the ephemeral worktree environment.

Closes #287