correct implementation of ndim ... (#40)

* correct implementation of ndim in order to work in compile mode with tensorflow (see issue tensorflow/tensorflow#48612 and #36)

* correct flake8

* Implement __len__ with shape for tf.function works

* import cast

eagerpy/tensor/base.py

@@ -51,7 +51,7 @@ def __bool__(self: TensorType) -> bool:
 
     @final
     def __len__(self: TensorType) -> int:
-        return len(self.raw)
+        return cast(int, self.raw.shape[0])
 
     @final
     def __abs__(self: TensorType) -> TensorType:
@@ -112,4 +112,4 @@ def __pow__(self: TensorType, exponent: TensorOrScalar) -> TensorType:
     @final
     @property
     def ndim(self: TensorType) -> int:
-        return cast(int, self.raw.ndim)
+        return len(self.raw.shape)