DyNet init function is not easy to use about dynet

[odashi]

I saw that currently there are 2 codeblocks to parse commandline args (one is in initialize(), and another is in initialize_gpu() which is called from the former).
They attempt to consume same args like --dynet-mem, and this behavior looks kind of confusing.
I think it might be better to integrate both to single codeblock as well as separating commandline parsing and actual configurations.

[odashi]

Oh thanks, new code look better!

[neubig]

Resolved by #117