chore(gallery): add Huihui Qwen3.8 Q5/Q6 builds

[localai-org-maint-bot]

Description

Add Q5_K and Q6_K GGUF builds of Huihui Qwen3.8 27B to the existing variant group. These provide intermediate quantizations between the existing Q4_K and Q8_0 builds. Both preserve the llama.cpp MTP settings, sampling defaults, and shared BF16 vision projector. Document explicit installation commands.

Notes for Reviewers

• Model source: https://huggingface.co/huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF/tree/b146b417080fd3e0e842e9539e69921707360aae
• All new download URLs are revision-pinned. SHA256 values and byte sizes for both weights and the projector match Hugging Face LFS metadata and independent HTTP HEAD x-linked-etag / x-linked-size responses.
• Parsed the full gallery and verified 2,085 unique names, valid non-nested variant references, matching weight/projector paths, unchanged Q8 runtime settings in the new builds, and 2,082 unrelated entries unchanged. git diff --check passes.
• go test ./core/gallery -count=1 cannot start in this environment: generated pkg/grpc/proto is missing, and protoc is unavailable. No inference test or full weight download was performed.
• Based on master at 1997c6cb832711db333de69c50b11913268a2bf7.

Signed commits

• Human DCO sign-off remains required; repository policy prohibits an AI from supplying it. The commit includes Assisted-by: Codex:gpt-6.
• Documentation updated (docs/content/) for user-facing changes