Add fiscal diagnostics comparison tool

[MaxGhenis]

Summary

• add a diagnostics-only tool for comparing two Populace US calibration_diagnostics.json files on a strictly shared target surface
• compute the same sqrt-value weighted, 50/50 amount/count, capped-MAPE objective used by the US fiscal refresh
• emit JSON, CSV, and Markdown artifacts with family contributions plus top candidate regressions/improvements
• reject duplicate diagnostic target keys and reject shared target names whose target values differ, so stale same-name comparisons cannot be treated as release evidence

Validation

• uv run ruff format tools/compare_us_fiscal_diagnostics.py packages/populace-build/tests/test_compare_us_fiscal_diagnostics.py
• uv run ruff check tools/compare_us_fiscal_diagnostics.py packages/populace-build/tests/test_compare_us_fiscal_diagnostics.py
• uv run python -m pytest packages/populace-build/tests/test_compare_us_fiscal_diagnostics.py

Real comparison guardrail

A previous candidate/incumbent smoke used diagnostics with overlapping target names but changed capital-gains target values. The tool now correctly rejects that comparison with ValueError: Cannot compare diagnostics with changed shared target values. The incumbent must be rescored on the exact candidate/current target surface before using this tool for release certification.