Add TensorBoard Support. And it has the following features:

• add a new parameter in train_options section, use_tensorboard.
• Set true to use tensorboard. It will record iteration error once every 25 iterations, epoch error once per epoch.
• There are tree types of error will be recorded. train_loss_iter is iteration loss, train_loss_last is the error of the last iteration in an epoch, train_loss_mean is the mean error of all iterations in an epoch.
• Learning rates are tracked as well.
• A folder named tensorboard_logs will be created in the working directory. Use tensorboard --logdir=tensorboard_logs to view the logs.

Fix #207