2. Make SwitchOrderLayer support for softmax activation
3. Fix bugs

* Fix  #2797
* It because trainer_config_helpers' __dfs_travel__ did not record the
  node which travelled, and if the topology has a recursive dependency,
  there are some nodes will be travelled multiple times.
* Add a `travelled` set to record which node is travelled.
* Also add a unittest for this situation.

1. Add switch function for switching image dimensions order
2. Add CpuMatrix::backwardSoftmax function
3. Add pixel softmax layer, python wrapper and grad_test

Fix issue #2562

For example:
layer.printer(input=[x, y], format="x=%s y=%s")

Also remove active_type argument for many layers in config_parser.py because it is automatically
handled by LayerBase.

No longer need to specify target_inlinks or is_seq.

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

User can provide an initializer in ParamAttr. The initializer will be called when at parameters.create().