Feat: Support latest Jumanji version

[WiemKhlifi]

What?

Upgrade to the latest Jumanji version of 1.0.1 instead of 0.3.1 and pin to the original and latest Jumanji and Matrax.

How?

• Change the requirements.txt to use original versions instead of a fork.
• Adapt all the wrappers and systems to use cached specs similar to jumanji wrappers and envs.

Extra:

• Note that this PR will be merged after pushing this PR into jumanji for Connector env updates.
• Note that in some wrappers, since the specs outputs will be cached, some attributes can't be retrieved if defined after the super().__init__(env) ( The self.__getattr__(env,name) in parent class can't get the attribute from env if it's defined with different name in the env wrapper class).
  For example:

# Should work:
super().__init__(env)
self.time_limit = self._env.time_limit 

# Shouldn't work:
super().__init__(env)
self.time_limit = self._env.max_episode_length

[RuanJohn]

Thanks for this @WiemKhlifi! Just a few questions, but it looks mostly good to me.
As a sanity check can you please do a few test runs to just check that the system performance is unaffected?

[sash-a]

Thanks Wiem, couple small things, mostly removing the git stuff from requirements.txt where possible

[RuanJohn]

Thanks @WiemKhlifi. Some suggestions from my side.

[RuanJohn]

I must admit I am not a massive fan of the not here. But I prefer it over having the conditional in the aggregation function. What do you think of just having the config option be aggregate_rewards instead of use_individual_rewards? Then we could change the conditional here to if self._aggregate_rewards.

Suggested change

	reward = aggregate_rewards(timestep.reward, self.num_agents, self._use_individual_rewards)
	if not self._use_individual_rewards:
	reward = aggregate_rewards(timestep.reward, self.num_agents)

[sash-a]

eh either way I prefer this 😄

[WiemKhlifi]

I chose the second option to use aggregate_rewards instead of the not which is less confusing 😅