RankedPairs: recorded tiebreak resolutions don't match the ranking actually used

[peterrrock2]

Summary

In RankedPairs._run_step, tied dominating tiers are resolved twice, inconsistently. The resolution stored in ElectionState.tiebreaks comes from tiebreak_set(tier, tiebreak=self.tiebreak), but the final ordering (and therefore who gets elected) is built separately with sort_candidates_pseudo_lexicographically, which ignores that resolution.

For tiebreak="lexicographic" (the default) the two happen to agree, so the bug is currently invisible.For any other setting, e.g. tiebreak="random", the reported resolution and the actual outcome can disagree.

Reproduction Code

from votekit.ballot import RankBallot
from votekit.pref_profile import RankProfile
from votekit.elections import RankedPairs

profile = RankProfile(
    ballots=[
        RankBallot(ranking=(frozenset({"A"}), frozenset({"B"})), weight=1),
        RankBallot(ranking=(frozenset({"B"}), frozenset({"A"})), weight=1),
    ]
)

# A and B tie head to head, so the final digraph has them in one dominating tier.
# Repeat a few times since the random resolution lands on (A, B) half the time.
for _ in range(20):
    e = RankedPairs(profile, tiebreak="random")
    state = e.election_states[-1]
    print(state.tiebreaks, state.elected)

The output then includes rounds like:

{frozenset({'A', 'B'}): (frozenset({'B'}), frozenset({'A'}))} (frozenset({'A'}),)

i.e. the state says the tie resolved with B first, but A was elected. Elected candidates are
always in lexicographic order regardless of tiebreak.