We now have Float[np.ndarray, ...] <: np.ndarray. Added basic torch tests. (#68)

This required quite a lot of refactoring! JAX supports virtual subclass registration (its metaclass is ABCMeta) but NumPy does not, so we have to actually subclass `np.ndarray`.
Simple stuff like __base__ hacking fails due to deallocator conflicts.

test/test_serialisation.py

@@ -1,4 +1,6 @@
 import cloudpickle
+import numpy as np
+import torch
 
 from jaxtyping import AbstractArray, Array, Shaped
 
@@ -6,5 +8,9 @@ from jaxtyping import AbstractArray, Array, Shaped
 def test_pickle():
     x = cloudpickle.dumps(Shaped[Array, ""])
     y = cloudpickle.dumps(AbstractArray)
+    z = cloudpickle.dumps(Shaped[np.ndarray, ""])
+    w = cloudpickle.dumps(Shaped[torch.Tensor, ""])
     cloudpickle.loads(x)
     cloudpickle.loads(y)
+    cloudpickle.loads(z)
+    cloudpickle.loads(w)