WIP KQ binary mask: make it a parameter, turn on via command line

It is a pain to implement binary mask to 32-bit value conversion on
NEON and AVX2, so I decided to make the binary mask optional

There is also a commented out (and not working) attempt for NEON
in this commit.

include/llama.h

@@ -340,6 +340,7 @@ extern "C" {
         bool embeddings;  // if true, extract embeddings (together with logits)
         bool offload_kqv; // whether to offload the KQV ops (including the KV cache) to GPU
         bool flash_attn;  // whether to use flash attention [EXPERIMENTAL]
+        bool binary_kq;   // whether to use binary KQ mask [EXPERIMENTAL]
 
         // Abort callback
         // if it returns true, execution of llama_decode() will be aborted