chore(gallery): add Xing4.0 GGUF variants

[localai-org-maint-bot]

Description

Add Q5_K_M, Q6_K, and Q8_0 builds of Xing4.0-29B-A4B to the existing Q4_K_M gallery entry's variants. All builds use the existing llama.cpp configuration, embedded chat template, and multi-token prediction settings.

The model appears in Hugging Face's trending list, but the gallery currently offers only its Q4_K_M build. No open Xing gallery PR duplicates these additions.

Pin all four downloads to revision 845d645bfa2be61b7774f2a12a8bec9e2231ec5d of the GGUF repository. Add the original model URL and correct the copied description, which calls the GGUF entry a Transformers release. Document direct installation of a variant.

Notes for Reviewers

Validation:

• Cross-checked all four SHA256 values from the Hugging Face ?blobs=true API against x-linked-etag headers at the pinned revision.
• Parsed the GGUF metadata and tensor tables for all four files using bounded HTTP range requests. Each declares xing4_0.nextn_predict_layers=1 and includes the six blk.40.nextn tensors.
• go test ./scripts/build/gallery -count=1 passed.
• go run ./scripts/build/gallery . gallery /tmp/localai-gallery-20261009-packaged passed.
• A local Go/YAML audit parsed all 2,086 entries, validated all 596 variant references, and checked the four Xing builds for unique names, matching backend settings, file paths, pinned URLs, and metadata checksums.
• git diff --check passed.

go test ./core/gallery -count=1 could not compile because generated pkg/grpc/proto files are absent. Downloading the repository's protoc release returned HTTP 403. Model inference was not run.

Signed commits

• Human DCO sign-off remains required. The commit carries Assisted-by, per repository policy.
• Documentation updated in docs/content/features/model-gallery.md.