gallery: add AMD Instella MoE 16B variants

[localai-org-maint-bot]

Summary

• add AMD Instella-MoE-16B-A3B-Think to the model gallery
• provide Q4_K_M and Q8_0 GGUF builds as host-selectable variants
• configure llama.cpp with the embedded Jinja tokenizer template and tool-parsing fallback
• record SHA256 values independently confirmed from Hugging Face LFS metadata and x-linked-etag headers

Validation

• git diff --check
• confirmed both artifact URLs resolve and their x-linked-etag values match the recorded SHA256 values
• go test ./core/gallery/... could not start locally because the environment returned HTTP 403 while fetching golang.org/x/sys@v0.0.0-20210809222454-d867a43fc93e from both the default proxy and direct source; CI is expected to run the gallery suite

Model sources

• https://huggingface.co/amd/Instella-MoE-16B-A3B-Think
• https://huggingface.co/DevQuasar/amd.Instella-MoE-16B-A3B-Think-GGUF