llama.cpp b10050 fixes Vulkan async copy race condition

The llama.cpp project has released version b10050, which includes a critical synchronization fix for the Vulkan backend. When using the transfer queue for asynchronous copies, the update ensures that an event_wait is performed to prevent race conditions.

• Resolves a race condition in Vulkan async copies by adding sync on event_wait (PR #25229).
• Disables KleidiAI support for macOS Apple Silicon builds.
• Provides binaries for CPU, GPU, and specialized accelerators across Windows, Linux, macOS, iOS, Android, and openEuler.

This release ensures stability for Vulkan users by addressing concurrency issues in data transfer operations.