[P14] Implement curse_score tool

[four-bytes-robby]

Goal

Implement the curse_score tool — ranks files by risk using the formula:
curse_score = changes × log2(authors+1) × exp(-0.5 × age_years) × log2(churn_rate+2) × acceleration

Parameters

• top (number, optional, default: 10) — number of files to return
• since (string, optional) — time window filter

Implementation

• Run git log --format='%H %an %ai' --numstat via ctx.$
• Parse output, compute per-file: change count, unique authors, recency, churn rate, acceleration
• Sort by score descending, return top N
• Exclude changelogs, lockfiles, CI config from results
• Return structured JSON with file paths, scores, and supporting stats

Tests

• Unit test with mocked git log output
• Verify curse formula computation
• Test since filtering
• Test exclusion of lockfiles/changelogs