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/runtime/runtime.h

@@ -99,7 +99,7 @@ void _intsort(int *, size_t);
 void _longsort(long *, size_t);
 bool _isheap(void *);
 int NtGetVersion(void) pureconst;
-unsigned _getcpucount(void) pureconst;
+int _getcpucount(void) pureconst;
 long _missingno();
 void __oom_hook(size_t);
 void _loadxmm(void *);