feat(idd-list): markdown-aware fence strip + cluster_leader config (cluster #14 #15)

[kiki830621]

Refs #14 #15

Summary

Cluster-PR addressing 2 idd-list spec follow-ups (rest of #14-#20 cluster — #16/#17/#19/#20 already shipped in v2.51.0 spec, retroactively closed; #18 wontfix-by-design):

• [refactor] idd-list PR body parsing: ignore #N inside fenced code blocks (sister concern from #13) #14: Step 3.5 PR body parsing 加 strip_fenced_code() helper — 移除 ... fenced blocks(支援 language tag）BEFORE regex scan,避免 PR body 含 example commit message 的 Refs #99 被誤當真實 ref。Inline code (`#99`) 不處理(false positive 罕見 + 會誤傷合法 inline refs)
• [enhancement] idd-list cluster_leader config (lowest | primary) (sister concern from #13) #15: Cluster leader 規則加 cluster_leader: "lowest" | "primary" config 選項 — lowest 為 default(維持 v2.51.0 行為,deterministic),primary 改用 PR body 中第一個 ref 順序。Invalid value fallback to "lowest"

Verification

Lightweight self-verify PASS:

• 純 spec language change(SKILL.md 是 prompt for Claude — 沒有 runtime code change to test mechanically)
• Pseudocode 完整 + Python imports 一致
• 對舊行為 backward compat:cluster_leader 缺省 fall back to "lowest" = v2.51.0 行為

Checklist

• Diagnose ✓ (cluster, both Simple — single-file spec change)
• Implement (1 commit, idd-list/SKILL.md +N lines)
• Verify ✓ (self-verify pseudocode consistency)
• Pending: human review of this PR + /idd-close after merge for both [refactor] idd-list PR body parsing: ignore #N inside fenced code blocks (sister concern from #13) #14 and [enhancement] idd-list cluster_leader config (lowest | primary) (sister concern from #13) #15

Cluster note

#14 and #15 share the same Step 3.5 spec section — single commit covers both. Per IDD cluster-PR mode, each issue still gets independent /idd-close summary.

---

🤖 docs/spec PR. Do NOT add 'Closes #14' / 'Closes #15' — IDD discipline.