Avoid using exec() for untrusted code execution

### 🔐 Security Issue: Avoid `exec()` for LLM-generated code

Currently, the system uses `exec()` to execute Python code returned by the LLM. This poses **severe security risks**.

#### 💡 Proposed Fix:
- Avoid `exec()`.
- Have the LLM return structured data (e.g., JSON filters) instead.
- Use Python’s native operators to apply the filter safely.

> Example: Instead of `df['Sales'] > 1000`, have the LLM return:
```json
{ "column": "Sales", "operator": ">", "value": 1000 }
````