add EN prompts for XCOPA (selected languages)

[haileyschoelkopf]

This PR adds EN prompts (from COPA) for the XCOPA languages that are in Bloom's training data, so we can use it for training Bloom-T0++ (name TBD :P)

@Muennighoff @thomasw21 @lintangsutawika

[afaji]

will add id translation

[Muennighoff]

Nice! 👻

1. There's a lot of double spaces. I didn't mark all, cuz maybe this is on purpose? As humans would still answer correctly.. (@VictorSanh, @stephenbach?)

2. Do we need to repeat the same template for each lang? If it's easiest, let's do it, else I think we can use something like @lintangsutawika described here & just use the same template for different langs.

[VictorSanh]

• There's a lot of double spaces. I didn't mark all, cuz maybe this is on purpose? As humans would still answer correctly.. (@VictorSanh, @stephenbach?)

Yes this kind of noise/variance is totally fine. Some prompts have a single \n, others have triple \n, and we encouraged this kind of variations.
Note that a handful of tokenizers have pre-normalization rules that remove these, but this is independent of the prompts

[VictorSanh]

lgtm, thanks! will merge as soon as the tests pass