Implement nnabla.experimental.distributions modules

[takuseno]

I implemented distribution modules such as chainer and tensorflow.

import nnabla as nn

from nnabla.experimental.distributions import Normal

loc = nn.Variable((2, 3))
scale = nn.Variable((2, 3))
inpt = nn.Variable((2, 3))
dist = Normal(loc, scale)

entropy = dist.entropy()
sample = dist.sample()
pdf = dist.prob(inpt)
variance = dist.variance()
mean = dist.mean()
std = dist.stddev()

For now, I've implemented three distributions.

• uniform distirbution
• normal distribution
• multivariate normal distrition

In testing, scipy.stats package is used as reference.

TODO

• determine sample interface
• unit test
• add comments
• add documents

Future works

• kl divergence
• consider more methods
• categorical distribution using random_choice Add random_choice function and tests. #395

cc @TE-TakuyaNarihira @TE-AkioHayakawa

[takuseno]

@TE-AkioHayakawa
I've created a squashed version of this PR at #565 . Please check it up :)