Fixing inefficiency in pyttb.tenrand. (#428)

The old version was creating random data as a multway arraw, reordering it to Fortran order (expensive and pointless), and then called the tensor constructor. Instead, we exploit the fact that we can just pass a 1D random array and then reshape that into the desired shape.

Co-authored-by: Danny Dunlavy <[email protected]>

Author: tgkolda

Diff for: pyttb/tensor.py

@@ -2959,8 +2959,7 @@ def tenrand(shape: Shape, order: MemoryLayout = "F") -> tensor:
     # Typing doesn't play nice with partial
     # mypy issue: 1484
     def unit_uniform(pass_through_shape: Tuple[int, ...]) -> np.ndarray:
-        data = np.random.uniform(low=0, high=1, size=pass_through_shape)
-        to_memory_order(data, order)
+        data = np.random.uniform(low=0, high=1, size=np.prod(pass_through_shape))
         return data
 
     return tensor.from_function(unit_uniform, shape)