Make general components

[pilgrimygy]

Put GaussianNetwork and DuelingNetwork into RLCore as general components. And make a general evaluate function.

[findmyway]

1. Remember to update NEWS.md.
2. The failure in CI seems to be unrelated to this PR. I'll take a look into it.

[findmyway]

It's quite strange that the tests now run much slower...

[pilgrimygy]

Thanks @findmyway for finding so many bugs!

[findmyway]

@pilgrimygy could you confirm that the JuliaRL_BasicDQN_CartPole experiment in this PR can be finished in less than a minute?

It seems everything works as usual on my local machine. Not sure why it is so slow in CI.

Now I get it. Will fix it soon.

[pilgrimygy]

Well. Why JuliaRL_BasicDQN_CartPole in CI is too slow?

[findmyway]

I must say, this is the weirdest thing I've seen in Julia so far. And I really don't know why...

(@v1.6) pkg> activate src/ReinforcementLearningExperiments/

(ReinforcementLearningExperiments) pkg> build

(ReinforcementLearningExperiments) pkg> test

┌ Warning: `InplaceableThunk(t::Thunk, add!)` is deprecated, use `InplaceableThunk(add!, t)` instead.
│   caller = ip:0x0
└ @ Core :-1

Then it is stuck. Just like what we see in the CI.

Well, if we run it in REPL:

(@v1.6) pkg> activate src/ReinforcementLearningExperiments/

julia> using ReinforcementLearningExperiments

julia> ex = E`JuliaRL_BasicDQN_CartPole`;

julia> run(ex)

Things work as usual.