fix: Update version of reverb & launchpad to work with distributed ppo.

Package updates to get distributed ppo working (probably other systems too):

Reverb v0.7.1 doesn't work with ppo (https://github.com/deepmind/acme/issues/223) due to a change in the rate limiters class. This is fixed in reverb v0.7.2.

Launchpad v0.5.1 doesn't work with the current distributed layout.

You get the following error:

program = get_agent_distributed(
File "/home/kaleab/Documents/Code/dynamic-networks/sparsity_search/agents/single_agent/single_agent.py", line 144, in get_agent_distributed
  program = ppo_jax.make_distributed_ppo(
File "/home/kaleab/anaconda3/envs/sparsity-rl/lib/python3.9/site-packages/acme/agents/jax/ppo/agents.py", line 83, in make_distributed_ppo
  return distributed_layout.make_distributed_program(
File "/home/kaleab/anaconda3/envs/sparsity-rl/lib/python3.9/site-packages/acme/jax/layouts/distributed_layout.py", line 346, in make_distributed_program
  replay_node = lp.ReverbNode(
TypeError: __init__() got an unexpected keyword argument 'checkpoint_time_delta_minutes'

Lp v0.5.2 has this param and thereby works.

google-deepmind / acme

fix: Update version of reverb & launchpad to work with distributed ppo. #224