feat: Merged PRs tracking, issue label filtering, AI summary, and commit activity stats

[Copilot]

Enhances the weekly report from a basic releases+issues dump into a high-signal digest of meaningful technical activity, with four new capabilities added to scripts/watch_dog.py.

New functions

• fetch_merged_prs() — GET /pulls?state=closed&sort=updated&direction=desc, filtered to merged_at > since; renders a 🔀 Merged PRs section (number, title, link, merge date, author)
• fetch_commit_stats() — GET /commits?since=…; renders a 📊 開發活躍度 section with total commit count and top-3 contributor breakdown
• generate_ai_summary() — aggregates releases, PRs, and issues across all watched repos into a prompt sent to gpt-4o-mini via https://models.inference.ai.azure.com using GITHUB_TOKEN; prepends a ≤150-character Traditional-Chinese summary to the report; fails silently

Modified

• fetch_issues() — accepts important_labels: list[str] | None; non-empty list filters to issues carrying at least one matching label; empty list/None preserves original show-all behaviour
• build_report() — collects all repo data upfront (required for cross-repo AI aggregation), then renders sections: AI summary → Merged PRs → Releases → Important Issues → Commit stats; new optional params important_labels and ai_summary
• main() — reads important_labels and ai_summary from config.json

Config additions

{
  "important_labels": ["bug", "enhancement", "breaking change", "priority/high"],
  "ai_summary": true
}

important_labels: [] disables filtering (backward-compatible default).

Dependencies

openai>=1.0.0 added to requirements.txt. The OpenAI import is wrapped in try/except ImportError so the module loads cleanly without the package (AI summary returns None).

Tests

19 new unit tests added (29 total); all existing tests updated to supply mock responses for the two additional API calls per repo (/pulls, /commits).