Dummy dataloader v2

[sg-wbi]

Supersedes #524

It adds a DummyKBExample to reduce boilerplate and sets examples per split, this way we can test the bypass args here #533.

An example is provided with dummy_kb_dataloader.py which implements an example which fails the "global_unque_ids" test.

[sg-wbi]

@hakunanatasha @galtay

[hakunanatasha]

@sg-wbi looks like this is fine; to run, I did the following:
python -m tests.test_bigbio tests/dummy/dummy_kb_dataloader.py and saw a failure.

The generator is very clever. Can we add in the tests case one for every sub function? This should be

• passage offsets
• entity offsets
• event offsets
• coreference
• q+a
• text2text
• text labels

this is a lot of work so i dont mind if its a TODO but this is good to merge as is imo

[hakunanatasha]

@sg-wbi are we merging this?

[sg-wbi]

If you think it looks good we can merge it!

Can we add in the tests case one for every sub function? This should be

passage offsets
entity offsets
event offsets
coreference
q+a
text2text
text labels

this is a lot of work so i dont mind if its a TODO but this is good to merge as is imo

You are right for completeness we should have these too but it will have to wait for now I guess...

[sg-wbi]

I am merging this as a first building block. I created this #665 so we do not forget about this.