* fix pruned model save and load, fix bugs of pruning depthwise conv

* add unit test of pruned model save and load

* temp delete

* add unit test

* add some unit tests

* complete autoprune unit test

* add sensitive unit tests

* fix some unit test

* fix auto_pruner.py

* Add group_param to collect relative covolution into groups.

* Support custom functions when computing pruned substructure of parameter
1. Support custom criterion of importance
2. Support custom importance sorting strategy