Text doesn't match results in multi-agent some_differences example

[EBoguslawski]

Environment

• Grid2op version: dev_multiagent branch
• System: windows
• python 3.12

Bug description

Text doesn't match expected output in the script examples/multiagent/some_differeces.py.

How to reproduce

Run examples/multiagent/some_differeces.py
or copy the following code:

import grid2op
from grid2op.multi_agent import MultiAgentEnv
from grid2op.gym_compat import DiscreteActSpace

if __name__ == "__main__":
    env_name = "l2rpn_case14_sandbox"
    centralized_env = grid2op.make(env_name)
    centralized_discrete = DiscreteActSpace(centralized_env.action_space, attr_to_keep=["set_bus"])
    
    ACTION_DOMAINS = {
        'agent_0' : [0, 1, 2, 3, 4],
        'agent_1' : [5, 6, 7, 8, 9, 10, 11, 12, 13]
    }
    
    ma_env = MultiAgentEnv(centralized_env, ACTION_DOMAINS)
    ma_discrete = {ag_id: DiscreteActSpace(ma_env.action_spaces[ag_id], attr_to_keep=["set_bus"]) for ag_id in ma_env.agents}
    print(f"There are {centralized_discrete.n} total unary actions on the centralized environment, but "
          f"(due to some lacking features) only {sum([el.n for el in ma_discrete.values()])} "
          f"for the decentralized version. "
          f"Basically, you cannot act on the \"interco\" at the moment")
    for sub_id in range(centralized_env.n_sub):
        agent_id = "agent_0"
        if sub_id in ACTION_DOMAINS["agent_1"]:
            agent_id = "agent_1"
        nb_cent = len(centralized_discrete.action_space.get_all_unitary_topologies_set(centralized_discrete.action_space, sub_id))
        
        loc_sub_id = ACTION_DOMAINS[agent_id].index(sub_id)
        nb_ma = len(ma_discrete[agent_id].action_space.get_all_unitary_topologies_set(ma_discrete[agent_id].action_space, loc_sub_id))
        print(f"sub_id  {sub_id}: {nb_cent} vs {nb_ma}")
    

Current output

There are ##179## total unary actions on the centralized environment, but (due to some lacking features) only ##180## for the decentralized version. Basically, you cannot act on the "interco" at the moment
sub_id  0: 3 vs 3
sub_id  1: 29 vs 29
sub_id  2: 5 vs 5
sub_id  3: 31 vs 31
sub_id  4: 15 vs 15
sub_id  5: 57 vs 57
sub_id  6: 4 vs 4
sub_id  7: 0 vs 0
sub_id  8: 15 vs 15
sub_id  9: 3 vs 3
sub_id  10: 3 vs 3
sub_id  11: 3 vs 3
sub_id  12: 7 vs 7
sub_id  13: 3 vs 3

Have a look at the numbers between ##

Expected output

There are ##179## total unary actions on the centralized environment, but (due to some lacking features) only ##N## for the decentralized version. Basically, you cannot act on the "interco" at the moment
sub_id  0: 3 vs n0
...
sub_id  13: 3 vs n13

We should have N < 179 and not N = 180 > 179

A lead: I think the do nothing action is counted twice when counting the multi-agent environment's actions.