feat: detect duplicate/similar commit subjects

[matmar10]

Summary

• Adds duplicate_threshold input (0.0–1.0, disabled by default) to action.yml
• Implements Sørensen–Dice bigram coefficient — no new runtime dependency
• Strips conventional-commit prefix (type(scope):) before comparing so feat(website): foo and feat(campaign): foo are correctly flagged as identical
• Handles double-colon typo (feat(auto-giving): : description) via + on the colon group
• Reports duplicate pairs in a new Duplicate Commit Subjects section of the PR comment, showing a markdown table with links and similarity percentage
• Increments the failure count so the action fails whenever duplicates exceed the threshold

How it works

1. After all commits are linted, every pair (i, j) is compared
2. normalizeSubject takes first line only, strips prefix, lowercases, collapses punctuation/whitespace
3. diceSimilarity computes the Dice coefficient over character bigrams
4. Any pair ≥ threshold is collected into duplicatePairs
5. core.setFailed is called if duplicatePairs.length > 0

Usage

- uses: matmar10/prcolinter@v1
  with:
    token: ${{ secrets.GITHUB_TOKEN }}
    duplicate_threshold: '0.8'

Test plan

• PR with feat(website): enable foo and feat(campaign): enable foo and threshold 0.5 → fails, shows duplicate section
• PR with unrelated commit messages → passes even with low threshold
• Omitting duplicate_threshold (default '') → feature disabled, no behaviour change
• duplicate_threshold: '1.0' → only exact-duplicate subjects after normalisation fail

🤖 Generated with Claude Code

[github-actions]

✅🙏🏻 Conventional Commit 🥳 🎉

🤖 Beep boop! Congrats, it like all your commit messages conform to the Conventional Commit spec! 👏👏👏

Your PR can be closed. Coffee is for closers, so here's a coffee for you: ☕️

Commit Message Lint Report

• ✏️ 5 commit(s)
• 👤 1 author(s)
• ❌ 0 lint error(s)
• ⚠️ 0 lint warning(s)