ggml : x2 speed for WASM by optimizing SIMD (ggml-org#11453)

* ggml : x2 speed for WASM by optimizing SIMD

* fix bad merging

* rm trailing spaces

* rm redundant clamp

* better quantize_row_q8_K

Co-authored-by: camel-cdr <[email protected]>

* remove memset that causes buffer overflow
Co-authored-by: camel-cdr <[email protected]>

---------

Co-authored-by: camel-cdr <[email protected]>