[Proposal] Space.seed should be reproducible for Space.seed

[Mark Towers]

Proposal

Space.seed will return a list of seeds, i.e., Discrete(3).seed(None) -> [1234]
However, we can't do Discrete(3).seed([1234]) to generate the same RNG

My proposal is the work code should work for all gymnasium spaces

seeding_values = space.seed(None) 
samples = [space.sample() for _ in range(3)]

space.seed(seeding_values)
new_samples = [space.sample() for _ in range(3)]

assert samples == new_samples

Motivation

One of the aims of Gymnasium is reproducibility, for space seed, a user must have a known seed for this to be possible currently as space.seed(None) does not guarantee that you get back all the information to recreate the space (particularly subspaces)

Pitch

Modify the necessary space.seed functions to enable this feature

Alternatives

This has worked in the past and no-one has asked for it yet

Additional context

No response

Checklist

• I have checked that there is no similar issue in the repo

[Mark Towers]

In private conversations with @RedTachyon, they clarified that they understood the Space.seed returns values to have this purpose and were surprised that this didn't work already.

Likewise, I don't understand how the current Space.seed return values are currently used, therefore, I think this is very reasonable API change to make

[Mark Towers]

Implemented within #1033