Fused delta-net (#1315)

* Revive fused delta-net

* Add command line argument for fused delta net

* Simplify/improve CUDA delta-net

* Add -fdn to llama-bench

* More CUDA fused delta net optimizations

* CPU optimizations

* Much faster fused delta-net on the CPU

It seems it is faster than the chunked implementation!

* Change meaning of fdn from bool flag to threshold value

* Use eps = 1e-6

* Give some nodes a name

include/llama.h

@@ -456,6 +456,7 @@ extern "C" {
         bool split_mode_graph_scheduling; // if true, force split mode graph scheduling
         //bool split_mode_f16;    // if true, cast intermediate results to f16 before copying to other GPUs
         bool scheduler_async;   // if true, with split mode "graph" graph evaluation will be done using multiple threads
+        int  fused_delta_net;
         bool mtp;   // Activate MTP if supported
         enum llama_mtp_op_type mtp_op_type;