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/catan.c

@@ -145,3 +145,7 @@ double complex catan(double complex z)
 	w = CMPLX(w, 0.25 * log(a));
 	return w;
 }
+
+#if LDBL_MANT_DIG == 53 && LDBL_MAX_EXP == 1024
+__strong_reference(catan, catanl);
+#endif