Fix VectorizeActionTransform for changing spaces

[pseudo-rnd-thoughts]

Description

Fixes #1169

gymnasium.wrappers.vector.VectorizeActionTransform was incorrectly implemented for changing spaces, i.e., self.same_out is False as the direction of data is the opposite of VectorizeObservationTransform from which it was copied.

[pkrack]

Question: why not self.single_action_space?
Docstring of iterate: "space: Observation space of a single environment in the vectorized environment."

Same docstring in concatenate which also takes as input a single_action_space. (also both docstrings are completely identical, the one for iterate is wrong).

This seems to suggest that iterate takes as argument the action space of a single action, not a batched one. Does it even matter since there is no out arg?

asking because I use it in my code, just want to make sure :)

[pseudo-rnd-thoughts]

I had just looked at the example docstring which shows a batched space rather than a single space so I'm using the batched space version.
I think the docstring is just incorrect, will update and check the rest.

[pseudo-rnd-thoughts]

Looks like someone, possibly me, copied the concatenate docstring for some reason, fixing now

[pkrack]

Thanks for the quick answers and fixes :)

I think in the end it does not matter whether the first arg is the batched space or not right? Both batched and unbatched versions of a same structured space should be identical up until the base spaces, which are iterated over just with iter regardless of which space is given as argument. The only difference it makes is with respect to typing, and that is only if someone uses literal shape annotations in the base spaces.

One more little comment before I stop bothering you:
The signature of iterate is:
def iterate(space: Space[T_cov], items: Iterable[T_cov])
but items is supposed to be a single batched sample from space ->
def iterate(space: Space[T_cov], items: T_cov)

[pseudo-rnd-thoughts]

For the second point, your correct, I'll update