Fix PerClassScorer overall totals deduplicating equal per-tag counts

[Chessing234]

Bug

PerClassScorer.get_metric computes the overall precision/recall/F1 by summing per-tag TP/FP/FN counts, but builds each sum from a set comprehension. Equal counts across tags collapse to a single element, so the overall totals (and the resulting precision/recall/F1) are understated whenever two or more classes share a count.

https://github.com/allenai/scispacy/blob/eacccd4/scispacy/per_class_scorer.py#L70-L79

sum_true_positives = sum(
    {v for k, v in self._true_positives.items() if k != "untyped"}
)
sum_false_positives = sum(
    {v for k, v in self._false_positives.items() if k != "untyped"}
)
sum_false_negatives = sum(
    {v for k, v in self._false_negatives.items() if k != "untyped"}
)

Example: with _true_positives = {'A': 5, 'B': 5, 'C': 3} the current code computes sum({5, 3}) = 8 instead of 5 + 5 + 3 = 13.

Root cause

The {} braces make each argument a set comprehension, which deduplicates equal values before sum() sees them.

Fix

Drop the braces so each expression is a generator expression passed directly to sum(); every per-tag count is now summed as intended. Behaviour is unchanged when all per-tag counts are distinct.