chore: bump llama.cpp to b9570; 1.0.9544:0 → 1.0.9570:0

[helix-nine]

Summary

Routine monitor-cycle upstream bump of llama.cpp.

• llama.cpp: b9544 → b9570 (latest fully-published ghcr.io/ggml-org/llama.cpp server images across all four variants — generic, cuda, rocm, vulkan; b9571–b9581 released but their container images aren't on ghcr yet).
• StartOS version: 1.0.9544:0 → 1.0.9570:0.
• start-sdk: already at latest (1.5.3) — no bump.
• npm update picked up patch bumps of devDeps @types/node (22.19.19 → 22.19.20) and prettier (3.8.3 → 3.8.4).

Upstream highlights (b9544…b9570)

• Adds Gemma 4 support (MTP, E2B/E4B assistants) and video input for multimodal models.
• Various GPU backend fixes (CUDA, Vulkan, HIP/RDNA3.5) and model-conversion fixes (Gemma4, Mistral-Medium-3.5).
• Internal server fix (#24281) — no llama-server flag renames, so the model presets in startos/actions/presets.ts are unaffected. No migration needed.

Full changelog: ggml-org/llama.cpp@b9544...b9570

Test plan

• npm run check (tsc --noEmit) — green.
• make/s9pk verification deferred to PR review per monitor-cycle policy.

[helix-nine]

Thanks for the review and merge, @MattDHill! 🚀 llama.cpp is now on b9570 (1.0.9570:0).