[Cui-yshoho]

What does this PR do?

Adds a dedicated llamafactory-qwen3-8-sft Claude Code skill for an end-to-end Qwen/Qwen3.8-27B workflow without changing the existing generic llamafactory-sft skill.

The skill covers:

• environment and dependency validation;
• data and model preparation;
• single-device 4-bit QLoRA;
• multi-device HyperParallel FSDP2 full SFT;
• smoke testing, held-out evaluation, export, and local API verification;
• explicit separation of code-confirmed, server-validated, and pending results.

Validation performed:

• skill structure validation (quick_validate.py);
• Qwen3.8 registration and template checks;
• schema parsing for both QLoRA and HyperParallel configurations;
• git diff --check.

Full model training remains an explicit runtime gate because it requires complete model weights, a compatible Ascend bitsandbytes build for QLoRA, and free NPU resources.

Before submitting

• Did you read the contributor guideline?
• Did you write any new necessary tests? (Skill-only change; validated against the existing Qwen3.8 registration and config parser.)