Random number generation should be consistent based on the simulation configuration

[tumaysem]

Random number generation should be consistent based on the simulation configuration.

All noise generation is based on randomness in simulation environment which can generate different output without a seed. value.

fixes #22

[coveralls]

Pull Request Test Coverage Report for Build 3792664276

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.0%) to 59.571%

---

Totals
Change from base Build 3641331801:	-0.0%
Covered Lines:	333
Relevant Lines:	559

---

💛 - Coveralls

[tumaysem]

I really wish to add the parameter to environment node in the config jSON file. But it seems there is a design decision to use the name of the node names as type of the environment to initialize. So I moved the random seed configuration to up one level.

"environment": {
  "NoisySharedSubgraph": {
      "noise": true
  }
},
"random_seed":42

[tumaysem]

I couldn't find any agent using pytorch randomization.