Separate quantized phi-3 implementation. (#2157)

* Separate quantized phi-3 implementation.

* Integrate the quantized phi3 model.=

* Small fixes, get the generation to work properly.

* Keep the old llama implementation around.

* Change the default.

candle-core/src/metal_backend/mod.rs

@@ -676,9 +676,6 @@ impl BackendStorage for MetalStorage {
             }
         }
 
-        if layout.is_contiguous() {
-        } else {
-        }
         Ok(Self::new(buffer, device.clone(), el_count, dtype))
     }
 

candle-core/src/sort.rs

@@ -178,7 +178,7 @@ impl crate::CustomOp1 for ArgSort {
             device.metal_device(),
             &command_buffer,
             kernels,
-            &name,
+            name,
             nrows,
             ncols,
             ncols_pad,