Observation normalization is not equivalent for `distributed` vs `threaded`

[rejuvyesh]

Dojo.jl/examples/reinforcement_learning/algorithms/ars.jl

Lines 119 to 127 in c513e78

	if distributed
	envs = [deepcopy(env) for i = 1:(2 * hp.n_directions)]
	normalizers = [deepcopy(normalizer) for i = 1:(2 * hp.n_directions)]
	hps = [deepcopy(hp) for i = 1:(2 * hp.n_directions)]
	print(" $(nprocs()) processors")
	else
	envs = [deepcopy(env) for i = 1:Threads.nthreads()]
	print(" $(Threads.nthreads()) threads")
	end

The distributed version updates a different normalizer for each run. Not sure if that was intentional.

[janbruedigam]

I removed parallelization in this example in the latest version to keep the examples rather simple. Parallelization would be a nice feature but I think it might make more sense to put that in its own package or in ReinforcementLearning.jl.
Feel free to reopen if you see it differently.