Add some more developed training examples. (#199)

* Use contiguous tensors for variables.

* Sketch the mnist example.

* Start adding the reduce ops.

* Renaming.

* Refactor the reduce operations.

* Bugfix for the broadcasting vectorization.

candle-core/src/dummy_cuda_backend.rs

@@ -40,7 +40,7 @@ impl crate::backend::BackendStorage for CudaStorage {
         Err(Error::NotCompiledWithCudaSupport)
     }
 
-    fn sum(&self, _: &Layout, _: &[usize]) -> Result<Self> {
+    fn reduce_op(&self, _: crate::op::ReduceOp, _: &Layout, _: &[usize]) -> Result<Self> {
         Err(Error::NotCompiledWithCudaSupport)
     }