Adding Boltzmann Softmax Exploration

[anmole17]

Please view the Contributing Guidelines for information on Contributing.

Is your feature request related to a problem? Please describe.
The Epsilon greedy method is not good at picking one of two high Q-values and it will always pick the highest Q-value and ignore all other high Q-values, this can sometimes lead to improper exploitation of all high valued actions.

Describe the solution you'd like
Implementing Boltzmann Softmax exploration, which helps Brain to pick the actions on a probabilistic manner, where all high Q values are given high probability, so each action has a probability of getting picked.

Additional context
Boltzmann Softmax Exploration