Fix imagenet models for v2

[mitmul]

Actually models other than NIN work correctly with the current code, but NIN model raises the error below with Chainer v2.

$ cd chainermn/examples/imagenet/models_v2
$ python -c 'from nin import NIN; model = NIN()'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/Users/shunta/Codes/chainermn/examples/imagenet/models_v2/nin.py", line 18, in __init__
    None, (96, 96, 96), 11, stride=4, wscale=w),
  File "/Users/shunta/.pyenv/versions/anaconda3-4.4.0/lib/python3.6/site-packages/chainer/links/connection/mlp_convolution_2d.py", line 69, in __init__
    argument.check_unexpected_kwargs(kwargs, wscale=msg)
  File "/Users/shunta/.pyenv/versions/anaconda3-4.4.0/lib/python3.6/site-packages/chainer/utils/argument.py", line 4, in check_unexpected_kwargs
    raise ValueError(message)
ValueError: wscale is not supported anymore. Use conv_init and bias_init argument to change the scale of initial parameters.

This PR fixes this and also update all other models for V2 style.

[iwiwi]

Thank you for the PR! We will shortly merge this after testing them locally.