[Docs] Add Unsloth fine-tuning tutorial for ERNIE-4.5-VL-28B-A3B (EN/ZH)

[mattheliu]

📚 Documentation Enhancement

This PR adds a complete fine-tuning guide for ERNIE-4.5-VL-28B-A3B using Unsloth,
with both English and Chinese versions to better serve the global community.

✨ What's Included

• Full end-to-end workflow:
  • Installation & environment setup
  • Model loading & LoRA configuration
  • Vision-language dataset preparation
  • Custom collator & trainer for ERNIE-4.5-VL
  • Training, inference, and model export
• Supports large-scale VL fine-tuning on limited GPU resources
• Covers production-grade deployment practices

📁 Added Files

• docs/source/Unsloth_ERNIE_4_5_VL_28B_A3B.md
• docs/source/Unsloth_ERNIE_4_5_VL_28B_A3B.zh.md

🎯 Impact

This documentation significantly lowers the barrier for developers and researchers
to fine-tune ERNIE-4.5-VL models efficiently and correctly, and provides
a reference-grade implementation for future multimodal projects.

[paddle-bot]

Thanks for your contribution!

[jzhang533]

LGTM