feat(essence): improve sentence scoring for better title/header retention

[amihos]

Summary

• Add first sentence bonus (+10) to preserve titles/headers during essence extraction
• Add markdown header detection and label-style headers (PROBLEM:, SOLUTION:)
• Add ISO date format detection (2025-11-29)
• Add more action verbs for better extraction (fixed, implemented, created, etc.)
• Ensure first sentence is always included if under 50% of max_len

Problem

The extract_essence() function was dropping important context like titles, headers, and section markers because they didn't score highly in the sentence scoring algorithm. This caused memories to lose critical retrieval context.

For example, a memory like:

2025-11-29: MUNygo Matching Algorithm - Product Thinking Session

PROBLEM: The matching logic was unclear
SOLUTION: Implemented asymmetric scoring...

Would have the title dropped entirely because it didn't contain action verbs, measurements, or other high-scoring patterns.

Test Plan

• Verify memories with titles/headers now retain their first sentence
• Verify markdown headers get high scores
• Verify label-style headers (PROBLEM:, SOLUTION:) get high scores
• Verify ISO dates (2025-11-29) get high scores
• Verify backwards compatibility - memories without headers still work

🤖 Generated with Claude Code

[nullure]

LGTM