feat: change coupling detection (co-change analysis)

[bntvllnt]

Summary

Detect files that frequently change together in git history but have no direct import relationship — hidden temporal coupling.

Motivation

• CodeScene's change coupling is one of their most valued features
• Files that always change together reveal hidden dependencies the import graph misses
• AI agents making changes need to know "if you touch X, you probably need to touch Y too"
• Amazon's AI code incident was partly caused by missing awareness of coupled systems

Approach

1. Parse git log --all --name-only (already done for churn)
2. For each commit, record which files changed together
3. Compute co-change frequency for file pairs
4. Filter by minimum threshold (e.g., changed together 3+ times)
5. Flag pairs with high co-change but no import edge (hidden coupling)

Proposed API Addition

Add to detect_changes or new get_coupling tool:

output: {
  temporalCoupling: Array<{
    fileA: string,
    fileB: string,
    coChangeCount: number,
    hasImportEdge: boolean,
    confidence: "high" | "medium" | "low"
  }>
}

Acceptance Criteria

• Co-change pairs computed from git history
• Hidden coupling flagged (co-change without import edge)
• Exposed via MCP tool
• Tests with real git repos

Priority

Short-term — High value, low effort (git data already parsed).