Sketch a fast cuda kernel for reduce-sum. (#109)

* Sketch a fast cuda kernel for reduce-sum.

* Sketch the rust support code for the fast sum kernel.

* More work on the fast kernel.

* Add some testing ground.

* A couple fixes for the fast sum kernel.

candle-core/examples/cuda_basics.rs

@@ -0,0 +1,15 @@
+#[cfg(feature = "mkl")]
+extern crate intel_mkl_src;
+
+use anyhow::Result;
+use candle::{Device, Tensor};
+
+fn main() -> Result<()> {
+    let device = Device::new_cuda(0)?;
+    let t = Tensor::new(&[[1f32, 2., 3., 4.2]], &device)?;
+    let sum = t.sum(&[0])?;
+    println!("{sum}");
+    let sum = t.sum(&[1])?;
+    println!("{sum}");
+    Ok(())
+}