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/acosh.c

@@ -31,7 +31,7 @@ asm(".ident\t\"\\n\\n\
 Musl libc (MIT License)\\n\
 Copyright 2005-2014 Rich Felker, et. al.\"");
 asm(".include \"libc/disclaimer.inc\"");
-/* clang-format off */
+// clang-format off
 
 /**
  * Returns inverse hyperbolic cosine of 𝑥.
@@ -53,3 +53,7 @@ double acosh(double x)
 	/* |x| >= 0x1p26 or nan */
 	return log(x) + 0.693147180559945309417232121458176568;
 }
+
+#if LDBL_MANT_DIG == 53 && LDBL_MAX_EXP == 1024
+__strong_reference(acosh, acoshl);
+#endif