WYL-52: suppress debater mention-chain cascade

Add _NO_MENTION_RULE to all DEBATER_PROMPTS and JUDGE_PROMPT_TEMPLATE
explicitly instructing agents not to @-mention other agents in their
replies. Mentions trigger Multica's mention-trigger mechanism, causing
cascade tasks from debaters responding to each other's comments.

Plain-name references (e.g. 'Code Reviewer' not '@Code Reviewer') are
still allowed for cross-reference in evidence text.

src/coordinator/orchestrator.py

@@ -48,6 +48,13 @@ JUDGE_NAME: str = "Reality Checker"
 # Per-debater evidence prompts.
 # Placeholders: {issue_title}, {issue_description}, {commit_url}
 # Each prompt tells the debater exactly which tool output to ground its argument in.
+_NO_MENTION_RULE: str = (
+    "**Important:** Do not @-mention any other agent or person in your reply. "
+    "If you need to refer to another debater, use their plain name only "
+    "(e.g. \"Code Reviewer\" not \"@Code Reviewer\"). "
+    "Mentions cause unwanted cascade tasks."
+)
+
 DEBATER_PROMPTS: dict[str, str] = {
     "Senior Developer": (
         "**Your task:** grep the committed code for LLM API calls "
@@ -55,26 +62,30 @@ DEBATER_PROMPTS: dict[str, str] = {
         "1. Yes/no — does the code make LLM API calls?\n"
         "2. The exact grep command you ran.\n"
         "3. The full grep output (or 'no matches').\n\n"
-        "Max 300 words."
+        "Max 300 words.\n\n"
+        + _NO_MENTION_RULE
     ),
     "Code Reviewer": (
         "**Your task:** diff the committed implementation method-by-method against "
         "every requirement in the issue description. For each requirement: state it, "
         "then label MET or MISSING with one line of evidence from the diff. "
         "Overall confidence: high / medium / low.\n\n"
-        "Max 300 words."
+        "Max 300 words.\n\n"
+        + _NO_MENTION_RULE
     ),
     "AI Engineer": (
         "**Your task:** ground-check every factual claim in the issue description "
         "against the committed code. For each claim: state it, then verify or refute "
         "with a direct code reference. Label each VERIFIED or UNVERIFIED.\n\n"
-        "Max 300 words."
+        "Max 300 words.\n\n"
+        + _NO_MENTION_RULE
     ),
     "Project Manager Senior": (
         "**Your task:** check the committed work against the original acceptance "
         "criteria. For each criterion: satisfied (with evidence) or not met. "
         "State your recommendation: ACCEPT or REJECT.\n\n"
-        "Max 300 words."
+        "Max 300 words.\n\n"
+        + _NO_MENTION_RULE
     ),
 }
 
@@ -96,7 +107,9 @@ Deliver your verdict. Begin your reply with exactly one of:
 - `VERDICT: ACCEPT` — implementation satisfies all critical acceptance criteria
 - `VERDICT: REJECT` — it does not
 
-Follow with reasoning (max 400 words).\
+Follow with reasoning (max 400 words).
+
+**Important:** Do not @-mention any agent or person in your reply. Use plain names only.\
 """
 
 _COMMIT_URL_RE = re.compile(