Make more ML improvements

- Fix UX issues with llama.com
- Do housekeeping on libm code
- Add more vectorization to GGML
- Get GGJT quantizer programs working well
- Have the quantizer keep the output layer as f16c
- Prefetching improves performance 15% if you use fewer threads

libc/tinymath/fsumf.c

@@ -22,8 +22,8 @@
 /**
  * Adds floats in array.
  */
-float fsumf(const float *p, size_t n) {
-  float s;
+double fsumf(const float *p, size_t n) {
+  double s;
   size_t i;
   if (n > 8) return fsumf(p, n / 2) + fsumf(p + n / 2, n - n / 2);
   for (s = i = 0; i < n; ++i) s += p[i];