Make the evaluators inside a recurrent goup true evaluator, meaning that their evaluation results are incorporated into the whole evaluator result.

1. v2.layer.parse_network does not correctly handle the generation output.
2. GatherAgentLayer does not correctly handle generation output when batch_size > 1
3. Fix CustomStackTrace for rnn group

No longer need to use SubsequenceInput. The framework will detect.

* also remove unused getValueStr

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Fix sparse training for trainer_count=1

For trainer_count=1, the gradient machine is NeuralNetwork, which does not create parameter buf for PARAMETER_GRADIENT for sparse update in Parameter::enableType. But gradient parameter buf is still used in SgdThreadUpdater.

* Minor update to comment

* hierarchical rnn document, add new config example

* update inputs_type of label

* add check for unsupported config

* refine hierarchical document

* refine doc title

* update docs, fix paddle to PaddlePaddle

* follow comments

* follow comments to fix bugs

* Correctly handling multiple inputs and integer inputs for recurrent_group

* Fix ScatterAgentLayer for generation

* Revert sequence_(nest)_rnn.conf

Sequences should be sorted according to the number of subsequences they have.

Change-Id: I8e0a8ea6fc2760652d9c76440a539c90860062d3