feat(gallery): add MiniCPM-V uncensored variants

[localai-org-maint-bot]

Description

Add AtomicChat's MiniCPM-V 4.7 35B-A3B Uncensored model with AD-Q4_K-IQ4_XS and AD-Q6_K variants for llama.cpp. Both builds include every weight shard and the F16 vision projector, use the embedded chat template, and default to 32K context. Add installation instructions under docs/content/.

This fine-tune is distinct from the original OpenBMB model proposed in #12686.

Notes for Reviewers

• Source: AtomicChat model card. All downloads are pinned to 948c2f37e34730f8abd9b80f86642f37e8f30960.
• Cross-checked SHA256 and size for all 28 unique files against both HF LFS metadata and pinned download HEAD headers.
• Verified LocalAI's pinned llama.cpp revision includes the required MiniCPM-V 4.7 support commit (69f201a).
• Strict YAML parsing and structural checks pass: 2,193 unique names, valid variant targets, complete 11/16-shard builds, matching model/projector paths, and all 2,191 existing entries unchanged. git diff --check passes.
• go test ./core/gallery/... cannot start: generated pkg/grpc/proto bindings are absent in the clean worktree. No inference test was run. The publisher also reports that vision was not tested after abliteration.
• Human DCO sign-off remains required; this AI-assisted commit contains only the Assisted-by attribution.

Signed commits

• Yes, I signed my commits.
• Documentation updated (docs/content/) for user-facing changes, or not applicable