Add wiki filter on "type" meta field

[SaulLu]

This PR add a filter that filters out all the examples that doesn't have their "type" field inside their "meta" value equal to "text".

I've tested it on lm_en_wikinews_filtered, here's the logs:

03/03/2022 11:46:20 - INFO - __main__ - Applied filter: filter_wiki_non_text_type
03/03/2022 11:46:20 - INFO - __main__ -      Initial number of samples: 54387 samples
03/03/2022 11:46:20 - INFO - __main__ -      Removed samples: 24736 samples
03/03/2022 11:46:20 - INFO - __main__ -      Removed percentage: 45.48 %

Partially solves #5

[HugoLaurencon]

Hey @SaulLu have you tested this code?

To me, iffilter_wiki_non_text_type is a function we should pass to the .filter method of datasets, it should take a single example as input, and return a boolean, not a list of booleans

Maybe your code works too, don't know

[lvwerra]

@HugoLaurencon I think it should work if .filter(some_filter, batched=True), no?

[lvwerra]

LGMT! 🚀

[SaulLu]

Remaining todo: have the list of datasets we want this filter apply on