add OpenZero Zero GGUF models to gallery

[ResearchForumOnline]

What changed

• adds zero-qwen3-8b-openzero-q5-k-m to the LocalAI gallery
• adds zero-gemma4-e4b-openzero-q5-k-m to the LocalAI gallery
• uses the appropriate Qwen3 and llama.cpp tokenizer-template configurations
• pins both Hugging Face downloads with verified SHA-256 hashes

Why

These entries make two standalone, one-file OpenZero GGUF releases installable through the LocalAI gallery. Both models target local coding, research, chat, and agentic workflows and were fine-tuned on 2,033 curated OpenZero examples.

User impact

LocalAI users can discover and install the Q5_K_M Qwen3-8B and merged Gemma4-E4B releases without manually creating model configuration files.

Validation

• parsed gallery/index.yaml successfully with PyYAML
• confirmed both new names occur exactly once
• ran git diff --check
• verified each Hugging Face x-linked-etag matches the committed SHA-256

[localai-org-maint-bot]

Reviewed the two gallery entries. The model names and file references are internally consistent, the Qwen3 and tokenizer-template configurations follow the existing gallery patterns, git diff --check passes, current master is an ancestor of this head, and DCO is green.

I could not independently re-fetch the Hugging Face x-linked-etag headers from this environment because its outbound proxy returned 403, and the focused go test ./core/gallery/... cannot start in a fresh checkout until the repository-generated pkg/grpc/proto package exists. Please let the normal gallery/CI checks run after marking the PR ready.

Looks good from review. @mudler