llama.cpp b9971 refactors server_stream and removes non-thread-safe rd.stop() call

The llama.cpp project released build b9971, which includes a refactor of the `server_stream` functionality. This update addresses thread safety by removing the non-thread-safe `rd.stop()` call and moving `server_res_spipe`. The release also updates developer documentation to reflect the improved API.

• Refactored `server_stream` and removed `spipe` from `server_http_res`
• Removed non-thread-safe `rd.stop()` call and moved `server_res_spipe`
• Updated dev docs for the improved server-stream API
• Disabled macOS Apple Silicon (arm64, KleidiAI enabled) builds
• Provided binaries for macOS, Linux, Android, Windows, and openEuler across CPU, GPU, and AI accelerator backends

The release ensures thread safety in the server streaming component while maintaining broad platform support.