docs: add RKLLM Rockchip NPU integration

[localai-org-maint-bot]

Description

This PR addresses #3192 by documenting the supported path for serving RKLLM models through LocalAI. It connects LocalAI's cloud-proxy backend to the OpenAI-compatible server shipped by upstream RKLLM, without redistributing Rockchip runtime binaries.

The guide covers:

• deploying the upstream RKLLM Flask server and a converted .rkllm model
• a complete LocalAI passthrough model configuration
• current upstream targets: RK3588, RK3576, RV1126B, and RK3562
• the current RK3566/Quartz64 limitation
• streaming, concurrency, API-surface, and tool-call limitations

It also adds RKLLM to the backend overview and compatibility table.

Notes for Reviewers

This is intentionally an integration through the existing wire-faithful proxy. Upstream already owns the SoC-specific runtime and OpenAI-compatible server; LocalAI adds its API surface, auth, routing, usage tracking, and UI in front of it.

Verification:

• git diff --check FETCH_HEAD...HEAD
• extracted YAML example parsed strictly and asserted for cloud-proxy/RKLLM fields
• go test ./backend/go/cloud-proxy
• independent review: no remaining findings

The full core/config suite reached 499/500 specs; its existing remote-fixture test was blocked by this environment's HTTP proxy fetching raw.githubusercontent.com.

Signed commits

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