Improve ML parsing reliability with fallback and calibration

[SahilKumar75]

Goal

Improve parser reliability when cell extraction is weak while preserving confidence calibration quality.

Scope

• Add fallback from cell parsing to text schedule parsing.
• Add confidence blending from cell and OCR quality scores.
• Add CLI support for parsing/tuning config files.
• Add tests for fallback and weighted confidence behavior.

Acceptance

• Worker can recover entries from raw text when cells fail.
• Confidence scoring remains stable and configurable.
• Tests pass.