Add argsort. (#2132)

* Add the argsort cuda kernels.

* CPU version of arg-sort.

* Hook the cuda kernel + rework the cpu bits.

* Add some dedicated test.

* Working cuda kernel.

* Metal kernel.

* Metal adjustments.

* Bugfix.

* Use the fast rope in qwen.

* Rework the expert selection in qwen.

candle-core/src/lib.rs

@@ -63,6 +63,7 @@ pub mod quantized;
 pub mod safetensors;
 pub mod scalar;
 pub mod shape;
+mod sort;
 mod storage;
 mod strided_index;
 mod tensor;