Unit-test bypass args

[hakunanatasha]

Supercedes #398

@galtay @sg-wbi

Adds the bypassing arguments to do the following:

(1) Ignore an entire data split (ex: ignore test)
(2) Ignore a key for all data splits (ex: ignore "entities")
(3) Ignore a SPECIFIC key in a SPECIFIC split (ex: Ignore "events" in the "test" split)

Split/key/split-key pairs will NOT affect the statistics function; I think it is important to see this either way.

The defaults for all of these are empty lists, meaning nothing is affected.

You can test them with the bionlp_st_2013_pc dataset that has the following statistics:

==========
id: 260
document_id: 260
passages: 260
entities: 7855
normalized: 0
events: 5992
coreferences: 455
relations: 0

validation
==========
id: 90
document_id: 90
passages: 90
entities: 2734
normalized: 0
events: 2129
coreferences: 128
relations: 0

test
==========
id: 175
document_id: 175
passages: 175
entities: 5312
normalized: 0
events: 0
coreferences: 0
relations: 0

Supported tasks include event extraction + coreference resolution, both which are missing in the test set.

You can test the script as follows:

# This will fail
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py 

# This passes
# Omit "events" AND "coreferences" in test
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py --bypass_split_key_pairs test,events test,coreferences

# This passes, but does not check test
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py --bypass_splits test

# This passes, but does not check events + coreferences
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py --bypass_keys events coreferences

# This will fail because the issue is with test not train
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py --bypass_splits train

# This fails because both events and coreferences are missing
python -m tests.test_bigbio biodatasets/bionlp_st_2013_pc/bionlp_st_2013_pc.py --bypass_split_key_pairs test,coreferences

@sg-wbi I'd like to add (or if you don't mind, you can add) some 'dummy' data to ensure this behavior is working as intended prior to merging. Otherwise, I have tested this on live data.

[galtay]

looks good to me!