use environment variable to set trainer

[jacquesqiao]

currently we use

paddle.init(xx1=yy1, xx2=yy2)

to init paddle, thus we need to change trianier python code if we want to change from local training to cluster training.

local init

paddle.init(use_gpu=False, trainer_count=1)

cluster training

# cluster init
paddle.init(
            use_gpu=False,
            trainer_count=1,
            port=7164,
            ports_num=1,
            ports_num_for_sparse=1,
            num_gradient_servers=1)

so we need to change this to config with environment variable，like:

export USE_GPU=True
export TRAINER_COUNT=1
...
python train.py

[typhoonzero]

I think this is resolve by #2014