Implement validation orchestration and issue prioritization

Inject evaluators via constructor to enable testing with fakes: BufdirFieldValidationService({ required CompletenessRuleEvaluator completenessEvaluator, required ThresholdComplianceRuleEvaluator thresholdEvaluator, required AnomalyDetectionRuleEvaluator anomalyEvaluator }). The validate() method should iterate report.sections, call each evaluator per section (passing prior section looked up by sectionId from the prior-period model), collect results into a flat list, then sort by severity ordinal. Define severity ordinals in the ValidationIssueSeverity enum: error = 0, warning = 1, info = 2. Use List.sort() with a comparator on severity ordinal.

Wrap each evaluator call in a try/catch and emit a synthetic ValidationIssue(severity: error, ruleClass: 'system', message: 'Validation error in section $sectionId: $e') on failure — this ensures one bad section never silently swallows other sections.

Unit tests using flutter_test. Test class: BufdirFieldValidationServiceTest. Required scenarios: (1) all fields valid, no prior → ValidationResult with empty issues and isValid true, (2) one completeness error → isValid false, error appears first in sorted result, (3) one threshold warning only → isValid true, (4) one anomaly warning only → isValid true, (5) mixed errors and warnings → errors sorted before warnings, (6) null prior-period model → anomaly evaluator runs without crash, (7) multiple sections with multiple issues → all issues aggregated, (8) empty BufdirReportModel (no sections) → empty ValidationResult, isValid true. Use test doubles (fake implementations) for each evaluator rather than real evaluators — this isolates the orchestration logic.

Minimum 90% line coverage.