[codex] Add congressional district outcome percentages

[MaxGhenis]

What changed

• adds district-level winner_percentage, loser_percentage, and no_change_percentage fields to the congressional district impact output
• computes outcome shares with the same +/-0.1% no-change threshold used in intra-decile results
• weights district outcome shares by household_count_people * household_weight, with a fallback when the people count column is absent
• adds focused tests for the new percentages, people-weighting behavior, and threshold handling

Why

The congressional district summary payload exposed average and relative income change, but not winner/loser shares. The frontend district map needs those shares in the state summary payload instead of computing them through hundreds of district-level requests.

Impact

US nationwide economy comparisons can expose district-level winner/loser/no-change shares directly in congressional_district_impact, which allows downstream clients to render those map modes from the main payload.

Validation

• uv run pytest --noconftest tests/test_congressional_district_impact.py was attempted, but the local build environment fails while compiling pydantic-core for Python 3.14 (Error: not inside a tmux session during Rust linking)
• the test file itself is updated and isolated, but I could not complete local execution because of that environment issue

Notes

• No policyengine-api-v2 route change is needed for this shape; the simulation service already returns the policyengine result blob directly.