[Bug] Wrong seed handling in DQN agent

[PabloHendriks]

Hi,

I was using the DQN implementation and noticed a bug with the seed handling.

rng = jax.random.PRNGKey(0)  # use a dummy rng here
 _action = basic_env.action_space().sample(rng)
_, _env_state = env.reset(rng, env_params)
_obs, _, _reward, _done, _ = env.step(rng, _env_state, _action, env_params)
_timestep = TimeStep(obs=_obs, action=_action, reward=_reward, done=_done)
buffer_state = buffer.init(_timestep)

This block of code (lines 81-86) used to initialise the environment overwrites the specified seed.

I have only worked with the DQN algorithm, so I do not know if the same issue applies to other agents.