POMDP value iteration algorithm

according to the Chapter two of 'Partially Observable Markov Decision Processes for Spoken Dialogue Management' written by Jason D. Williams.

Here we implement the value iteration algorithm for POMDP with an example of voicemail.

There are two versions of code. Most of them are the same, the only difference is that the matrix_version use matrix multiplication instead of for-loops when updating values using bellman equation: 

v^{a,k}(s) <-- r(s, a) + gama * sum_{s'} sum_{o'} P(s'|s, a) * P(o'|s', a) * V^{k(o')}_{t-1}(s')

For convenience, the initialization, value iteration and pruning functions are written in three different file and called in main.py