- Action collects shared operations: data_forward, mode, pad, make_batch
- Trainer and Tester receives Action as a parameter
- seq_labeling works in such setting

A lot of changes.

- [action] add k-means bucketing, partition sequences into buckets of nearly the same length
- [trainer] print train loss every 10 steps
- [loader] cws pku loader split sequence longer than max_seq_len into several shorter sequences

- [model] optimize cuda support in seq labeling model
- [test] add test data "pku" for chinese word seg
- test_tester.py and test_cws.py is OK to run!

- [tester][trainer] add cuda support
- [preprocess] fix label2index for padding label seq
- update README.md
- [test] add test_tester.py
- rename "action" to "core"

- [trainer]rename "batchify" to "make_batch" in trainer
- [trainer]pack (batch_x_pad, seq_len) into batch_x in make_batch for seq labeling, because seq length before pad is needed to make masks
- [trainer]unpack it in data_forward
- [model]shorten model definition
- [inference]build inference class. test_POS_pipeline.py is OK to infer
- [preprocessor]handle pickles in a nicer manner
- [FastNLP] add fastNLP.py as high-level API, not finished yet

# Conflicts:
#	fastNLP/modules/encoder/embedding.py
#	fastNLP/modules/encoder/lstm.py

add default parameters for modules, in order to decrease required params; update POSPreprocessor to return multi-level lists; add metrics.py

Add classification model.

optimize evaluation output in Trainer;
keep POS pipeline (loader + trainer + tester + saver) OK;
add codes borrowed from FudanParser.

update config loader

add ConfigSection class to save config data

restructure module: 4 classes; add modules; move prototype and rename

update config loader