fix: clarify semantic question prompts to output only structured JSON

Changed final answer instructions from
'When you have enough evidence (STRUCTURED OUTPUT):'
to 'When you have enough evidence, respond with ONLY this JSON (no reasoning, tool_call, or is_final fields):'

Prevents LLM from outputting both wrapper format and structured format.
Applied consistently across all 4 tiers (TERSE, BALANCED, DETAILED, EXPLORATORY).

Expected output:
{"analysis": "...", "evidence": [...], "related_components": [...], "confidence": 0.95}

Not:
{"reasoning": "...", "is_final": true}
{"analysis": "...", ...}

Author: jruokola

crates/codegraph-mcp/src/semantic_question_prompts.rs

@@ -33,7 +33,7 @@ RESPONSE FORMAT (JSON):
   "is_final": false
 }
 
-When you have enough evidence to answer (STRUCTURED OUTPUT):
+When you have enough evidence to answer, respond with ONLY this JSON (no 'reasoning', 'tool_call', or 'is_final' fields):
 {
   "analysis": "Direct answer to the question based on graph evidence gathered. Cite specific node IDs and relationships found.",
   "evidence": [
@@ -118,7 +118,7 @@ RESPONSE FORMAT (JSON):
   "is_final": false
 }
 
-Final answer format (STRUCTURED OUTPUT):
+When you have enough evidence to answer, respond with ONLY this JSON (no 'reasoning', 'tool_call', or 'is_final' fields):
 {
   "analysis": "Comprehensive answer based on graph evidence. Structure: 1) Direct answer 2) Supporting evidence from tools 3) Specific node IDs and relationships found 4) Confidence level and any limitations",
   "evidence": [
@@ -239,7 +239,7 @@ RESPONSE FORMAT (JSON):
   "is_final": false
 }
 
-Final answer format (STRUCTURED OUTPUT):
+When you have enough evidence to answer, respond with ONLY this JSON (no 'reasoning', 'tool_call', or 'is_final' fields):
 {
   "analysis": "COMPREHENSIVE ANSWER with structure:
 
@@ -437,7 +437,7 @@ RESPONSE FORMAT (JSON):
   "is_final": false
 }
 
-Final answer format (STRUCTURED OUTPUT):
+When you have enough evidence to answer, respond with ONLY this JSON (no 'reasoning', 'tool_call', or 'is_final' fields):
 {
   "analysis": "EXHAUSTIVE MULTI-DIMENSIONAL ANSWER: