Change discrete dtype to np.int64

[pseudo-rnd-thoughts]

The Discrete class definition return type is int while the space dtype is np.int64.
This PR makes the Discrete type consistent with np.int64.

[younik]

Side question: do we need np.int64 instead of np.int32?

I don't have an example in mind where you need int64 for Discrete; while usually int32 is used

[pseudo-rnd-thoughts]

Agree, this was my initial thoughts as well but the numpy random integers and choice functions return np.int64 and the dtype was np.int64 already.

[younik]

Agree, this was my initial thoughts as well but the numpy random integers and choice functions return np.int64 and the dtype was np.int64 already.

If we want to do this, we can change it by passing dtype to integers and an explicit arange to choice

[pseudo-rnd-thoughts]

@younik I have decided not to change it for three reasons

1. When you convert python int to numpy, its type is np.int64
2. numpy functions default dtype for integers is int64
3. MultiDiscrete uses np.int64

Therefore, Im going to keep it to np.int64 and it minimise the difference for backward compatibility.

[RedTachyon]

Can't we use one of the general numpy dtypes to allow int/int64/int32/whatever other int? The annotation could be int | np.integer, then I think it would accept int, int64, int32, uint8 and all of that stuff. While it does make sense to narrow down the return type, this would throw redundant errors (since all of those can be cast to int64 anyways)

[pseudo-rnd-thoughts]

Yes, I can edit to use np.integer[Any] and this should cover all of the cases mentioned

[RedTachyon]

Also in L24

[pseudo-rnd-thoughts]

Done