feat(gallery): add Spark 1.7B BF16 variant

[localai-org-maint-bot]

Description

Add the official Spark-X2.5-1.7B BF16 GGUF as a variant of the existing Q4 entry. Hosts with enough memory can select BF16 automatically; users can choose Q4, Q8, or BF16 explicitly. The new entry uses the existing llama.cpp configuration, embedded Jinja template, and 32,768-token context.

The 3.42 GB weights are pinned to Hugging Face revision 06c93e4e998a8e53e39b7b53d9bb842c9fbb5674. The gallery guide now includes the BF16 install command and explains how to keep a smaller build.

Sources: original model, official GGUF repository.

Notes for Reviewers

• Cross-checked SHA256 against pinned Hugging Face LFS metadata and the download's x-linked-etag header: 67d5f2f06e6d898efcf0dc40cab8528bc82b871c8dafb0936784183d2c10cdd9.
• Read the GGUF header through an HTTP range request: architecture spark2_5, file type 32 (BF16). Verified both in the llama.cpp revision pinned by LocalAI.
• go test ./core/gallery -count=1 -ginkgo.focus='gallery/index.yaml|gallery variant lint helpers' passes before and after the change.
• Parsed the complete YAML and verified the new model path, variant link, and that unrelated entries are unchanged.
• git diff --check passes.
• Full go test ./core/gallery -count=1: 392 passed, 3 failed, 9 skipped. The failures fetch external GitHub/Gist fixtures that return Forbidden in this environment.
• Inference was not run. No backend changes are required.

Signed commits

• Human DCO sign-off remains required; the commit records AI assistance only.
• Documentation updated under docs/content/.