Add the gather op. (#219)

* Start adding gather.

* Gather cpu implementation + use in simple training.

* Add scatter_add for the gradient of gather.

* Simple cpu implementation of scatter_add.

* Use gather in the simple-training backprop.

candle-core/src/dummy_cuda_backend.rs

@@ -85,6 +85,22 @@ impl crate::backend::BackendStorage for CudaStorage {
     fn index_select(&self, _: &Self, _: &Layout, _: &Layout, _: usize) -> Result<Self> {
         Err(Error::NotCompiledWithCudaSupport)
     }
+    fn gather(&self, _: &Layout, _: &Self, _: &Layout, _: usize) -> Result<Self> {
+        Err(Error::NotCompiledWithCudaSupport)
+    }
+
+    fn scatter_add(
+        &self,
+        _: &Layout,
+        _: &Self,
+        _: &Layout,
+        _: &Self,
+        _: &Layout,
+        _: usize,
+    ) -> Result<Self> {
+        Err(Error::NotCompiledWithCudaSupport)
+    }
+
     fn index_add(
         &self,
         _: &Layout,