llama.cpp b9277 moves save-load-state tests to proper test suite

llama.cpp, the leading open-source library for running large language models locally on consumer hardware, released version b9277 with a focused quality-of-life improvement for developers. The key change is the migration of the save-load-state functionality from a standalone example (in examples/save-load-state/) to a formal test suite (tests/test-save-load-state.cpp). This refactoring includes removing the old directory from the example CMakeLists and adding the new test to the testing framework as a 'model test'.

While seemingly minor, this move signals the project's maturation and commitment to long-term maintainability. By converting a demo into a repeatable, automated test, llama.cpp ensures that state-saving and loading (critical for checkpointing model runs) remains robust across all supported hardware. The release also updates the CODEOWNERS file and provides CI builds for a wide range of architectures including Apple Silicon, Intel, Vulkan, CUDA, ROCm, and ARM64. For the 112k+ developers following the project, b9277 reduces the risk of regressions and makes it easier to contribute without breaking core functionality.