Bug: improvement_detect_confusion_patterns conflates two data sources

Bug Description

The improvement_detect_confusion_patterns MCP tool returns confusing/misleading output by mixing two different data sources without making it clear.

Current Behavior

Tool returns:

{
  "total_events": 27,
  "resolved_events": 6,
  "unresolved_events": 21,
  "resolution_rate": 100
}

This is contradictory - if 6/27 are resolved, resolution rate should be ~22%, not 100%.

Root Cause

After investigation, the tool appears to combine:

1. improvement_confusion_events table: 6 events, 6 resolved = 100% resolution ✓
2. user_correction observations: 21 records (no matching confusion event records)

The "27 events" = 6 + 21, but only the 6 from the dedicated table have resolution tracking.

Expected Behavior

Either:

Option A: Separate the metrics clearly

{
  "confusion_events": {
    "total": 6,
    "resolved": 6,
    "resolution_rate": 100
  },
  "user_corrections_without_confusion_event": {
    "total": 21,
    "note": "These corrections don't have matching confusion_event records"
  }
}

Option B: Only report from the dedicated table

{
  "total_events": 6,
  "resolved_events": 6,
  "resolution_rate": 100,
  "note": "From improvement_confusion_events table only"
}

Impact

• D10 (Confusion Removal) dimension scoring uses this tool
• Misleading output can cause incorrect dimension scores
• Auditors may trust the numbers without realizing the data source mixing

How Discovered

During /improve self-audit, questioned why 6/27 = 100% resolution rate. Direct database queries revealed the two-source conflation.

Files to Check

• ecosystem-mcp/src/nautical_ecosystem/tools/improvement.py (likely location of detect_confusion_patterns implementation)