ggml : x2 speed for WASM by optimizing SIMD

A recent pull request by Xuan-Son Nguyen has significantly improved the speed of WASM by leveraging SIMD instructions for specific dot product functions. Most of the code improvements were made by the AI model DeepSeek R1, while Xuan-Son focused on developing tests and writing prompts. The author highlights promising results from DeepSeek R1 in optimizing code for machine learning models.