Make network op

Fix bug in SequenceSoftmax

Also remove operator bool in Error. The Error should be removed
later because it is not necessary for Paddle. We are now using Enforce
to handle error.

* default learning rate, temperary fix

* update

enable v2 automatically sets using cudnn-batch norm.

"fix gen list"

Make PADDLE_ENFORCE and PADDLE_THROW catchable

Feature/middle level net api

Make C lib in `paddle.v2.master.client` lazy load

* Use EnforceNotMet to unify all exception types.

Add ImageExpandFunction.

Make implement an operator less noisy.

Fix hl_sequence_avg_forward.

Currently, WITH_GOLANG is a compiling flag, and not all Paddle binary
enable `WITH_GOLANG=ON`. So we cannot ensure that every paddle has
`master.so`. Make load that C libary lazily could let the user run
Paddle without `WITH_GOLANG=ON`.

fix conv layer reshape

add param_attr parameter for img_conv_groups(...) in /python/trainer_config_helpers/networks.py

update gradient operator registry mechanism

Implement distributed training save model, improve master.NewClient i…

do not do log.Errorln when checkpoint is not found (which is normal)

OLD: op_type -> grad_op_creator

NEW: grad_op_type -> grad_op_creator
     op_type -> grad_op_type

Backward