Fast Hadamard Transform (#1249)

* Working hadamard for powers of 2

* working for m*2^k

* add scale and check contiguity

* add size check

* clean up

* fix test

* add grads + vmap

* gpu only

* skip on linux

* test typo

* add cpu impl

* remove gpu only tests

* fix linux build + add is_equivalent

mlx/ops.h

@@ -131,6 +131,12 @@ array flatten(
 /** Flatten the array to 1D. */
 array flatten(const array& a, StreamOrDevice s = {});
 
+/** Multiply the array by the Hadamard matrix of corresponding size. */
+array hadamard_transform(
+    const array& a,
+    float scale = 1.0f,
+    StreamOrDevice s = {});
+
 /** Remove singleton dimensions at the given axes. */
 array squeeze(
     const array& a,