feat(llm): improve integration prompt to always suggest placements

[twaugh]

Enhanced the Phase 3 integration prompt to ensure the LLM always provides at least one suggestion, and preferably multiple (2-3) suggestions when there are valid alternatives.

Changes to System Prompt:

1. Added explicit requirement to ALWAYS output at least one decision
2. Set preference for 2-3 decisions when multiple valid placements exist
3. Removed hard confidence threshold - if all candidates < 0.30, still output the best match with its actual confidence score
4. Reorganized prompt structure with reasoning-first format (matching classification prompt pattern)
5. Added detailed reasoning guidelines with 4-step decision framework
6. Changed system prompt role from 'JSON-only integration planner' to 'knowledge organization expert specializing in Logseq integration'. This leads with expertise/capability rather than constraints, creating a stronger professional identity that may improve LLM performance.

Changes to User Prompt (Examples):

1. Replaced "no output" examples with positive examples showing:
   • Best match even with weak semantic fit (confidence 0.55)
   • Multiple suggestions for ambiguous content (3 suggestions)
2. Expanded examples from 5 to 7 with richer reasoning chains
3. Added "CRITICAL REMINDERS" section reinforcing key requirements
4. Structured with and tags for better attention

Expected Benefits:

• User always gets at least one placement suggestion to consider
• Multiple options shown when knowledge could fit in several places
• More transparent reasoning helps user understand LLM decisions
• Consistent with classification prompt's chain-of-thought approach
• Better handling of edge cases (new topics, weak matches)

Assisted-by: Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84.59%. Comparing base (d200961) to head (e6192ef).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #35   +/-   ##
=======================================
  Coverage   84.59%   84.59%           
=======================================
  Files          48       48           
  Lines        5084     5084           
=======================================
  Hits         4301     4301           
  Misses        783      783           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.