llama.cpp b10452 refactors chat content handling for better multimodal support

The ggml-org team pushed llama.cpp b10452 to the project's 124k-star repo on August 16. This incremental release focuses on cleaning up the chat backend: a refactor of supports_string_content and supports_typed_content (PR #27130) simplifies how the server handles different content types in chat messages, which is critical as multimodal models become more common. It also improves capability detection for string content and adds a 'skip' option to the messages_inp_normalizer test. While not a flashy feature drop, the changes reduce edge-case bugs and make the codebase more maintainable, ensuring llama.cpp remains a reliable foundation for local LLM deployments.

As usual, b10452 ships an impressive matrix of prebuilt artifacts—dozens of combinations covering CPU, Vulkan, CUDA 12/13, ROCm 7.14, OpenVINO, SYCL, and even OpenCL Adreno for Android. The release includes the llama.app website link and UI assets, reflecting the project's push beyond CLI. For developers running self-hosted chat interfaces, these tweaks mean fewer surprises when mixing text and image inputs. The steady flow of weekly snapshots keeps llama.cpp at the forefront of edge inference, with no signs of slowing down.