Tools expect samples from a space to be a consistent size

Initially for actions I've introduced and used the `DiscreteParametric(space)` type, but samples from this space
are not constant sized which means ML/RL frameworks have a hard time when trying to allocate tensors/arrays for rollouts.

```python
>>> space = DiscreteParametric(3, ([Discrete(2), Discrete(3), Tuple([Discrete(2), MultiBinary(4), Discrete(2)])]))
>>> for i in range(10):
... print(space.sample())

(0, 1)
(0, 0)
(0, 1)
(0, 0)
(1, 2)
(2, (1, array([0, 1, 1, 0], dtype=int8), 0))
(2, (1, array([1, 1, 0, 1], dtype=int8), 0))
(0, 0)
(1, 2)
(2, (1, array([1, 1, 1, 1], dtype=int8), 0))
```

Perhaps a better idea is a constant sized, zero padded action space.