Revert to Mistral-7B-Instruct-v0.3 for local development

Reverted from SmolLM3-3B back to Mistral-7B-Instruct-v0.3 which works
correctly for local development. SmolLM3 had issues with JSON parsing
that caused tool calls to be displayed instead of executed.

Mistral-7B works locally but is not available on HuggingFace Inference
API free tier, so production deployment will require account upgrade
or alternative deployment method.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

app.py

@@ -126,10 +126,9 @@ async def query_mistral_async(message: str, language: str = "en") -> dict:
             {"role": "user", "content": f"Language: {language}\nQuestion: {message}"}
         ]
 
-        # Call LLM via HuggingFace Inference API (free tier)
-        # SmolLM3 is one of the few models available on hf-inference for chat
+        # Call Mistral via HuggingFace Inference API
         response = client.chat_completion(
-            model="HuggingFaceTB/SmolLM3-3B",
+            model="mistralai/Mistral-7B-Instruct-v0.3",
             messages=messages,
             max_tokens=500,
             temperature=0.3