feat: add statewide election probabilities

[dlworrell]

Summary

Adds the supplied Washington All Results workbook as the dated statewide snapshot, extends the standard-library importer for native XLSX, and regenerates the EDT election reports.

The report now gives every modeled decision boundary:

• a conditional tempered beta-binomial probability of change before certification;
• a separate 0–100 evidence-reliability score and grade;
• a probability-sorted at-a-glance table; and
• machine-readable model, forecast, threshold, and reliability fields.

The statewide remaining-ballot forecast is explicitly documented as a historical proxy (40.43% official 2022 primary turnout applied to about 5.1 million current voters), with forecast evidence set to 45/100. King County retains its official 45% turnout forecast with evidence 75/100.

Source workbook SHA-256: fd4d2a14a01d3945095e92cbb0d0a85b1c0fadc025882c73ed9348da1fbdf43a.

AES-DEV-001

• Governing architecture, specification, or ADR is linked or updated.
• Documentation is updated in this change series.
• Tests are included.
• Interface and versioning impact is addressed: tool version 0.2.0 and analysis schema version 2.
• Failure modes, recovery, observability, and diagnostics are documented.
• The change is one logical, reviewable unit suitable for revert and git bisect.

Governing design: docs/engineering/ADR-001-election-result-reporting.md.

AES-SEC-001

• The uploaded workbook is treated as untrusted external input.
• XLSX archive size, relationships, shared-string indices, schemas, numeric values, and normalized contest uniqueness are validated.
• No banned unsafe interface or custom cryptography is introduced.
• Unsafe operations are not introduced.
• Parser regression and publication evidence is supplied.
• No waiver is required.

The XLSX reader uses only Python's standard library and rejects archives above a 50 MiB uncompressed limit. Source hashes remain in the generated evidence.

Evidence

• 11 unit and integration tests pass on Python 3.12.
• 208 races generated from 2 immutable snapshots.
• 153 modeled boundaries have numeric probabilities; 60 races without a modeled boundary are explicitly marked not applicable.
• 213 at-a-glance rows cover all races and separate judicial boundaries.
• No race has valid votes exceeding its reconciled ballot denominator.
• Seven source denominator discrepancies are disclosed and reconciled from valid votes + overvotes + undervotes.
• EDT overall quality: 100/100, publication ready.
• EDT validation: 0 findings, 0 errors, 0 warnings.