Fix MajAtN IndexError when the gold is not the first choice

[iamsharduld]

What

MajAtN.compute builds the scoring Doc from the preprocessed golds:

processed_choices = [self.preprocess(text=g) for g in doc.get_golds()]   # length == #golds
new_doc = Doc(choices=processed_choices, query=doc.query, gold_index=doc.gold_index)

but keeps the original gold_index. When the gold is not the first choice, new_doc.get_golds()
indexes choices[gold_index] beyond the gold-only list and raises IndexError:

doc = Doc(query="q", choices=["London", "Paris", "Berlin"], gold_index=[1], task_name="test")
MajAtN(n=3).compute(doc, ModelResponse(text=["Paris", "Paris", "London"]))  # IndexError
# PassAtK on the same doc returns a correct score

Fix

Mirror PassAtK / GPassAtK: preprocess doc.choices and keep gold_index, so new_doc is
internally consistent for any gold_index.

Tests

tests/test_unit_base_metrics.py::test_maj_at_n_non_first_gold — majority-correct → 1, majority-wrong
→ 0, with gold_index=[1]. Fails before (IndexError), passes after.