Implemented Evidence Analysis for Lightweight Input Observation

Implemented an evidence-analysis foundation that aggregates signals from lightweight input observation and determines which regions should be passed to later, deeper analysis.

Multiple observations can be related and organized into boundary and candidate-region information while preserving traceability back to the supporting evidence for downstream processing.

This moves the Input Foundation toward selecting relevant regions from observed evidence instead of applying expensive analysis across the entire input.