Add more templates to show typical usage of Confidential AI and TEE

[h4x3rotab]

• Fine tuning template in GPU TEE: unsloth

[renezander030]

I'd like to pick up the unsloth item from this list.

Scope I have in mind — checking here before writing any code:

• New templates/unsloth-finetune/ following the existing template layout (docker-compose.yml, README.md, .env.example), registered in config.json and passing validate.py
• QLoRA fine-tune of a small open model (e.g. a 7–8B Qwen/Llama) with unsloth on a sample dataset, running in a GPU TEE CVM, adapters saved to the CVM volume
• README covering deploy, swapping model/dataset via env vars, and verifying the attestation of the training environment

Two questions before I start:

1. @h4x3rotab is this still the wanted shape, or has the direction moved since December?
2. @Leechael any constraints on base image, target GPU, or how a template gets into the prebuilt listing?

I'll only start after a go-ahead.