refactor: replace mock reasoning with real LLM-based engine

- Rewrite core.py: all mock implementations replaced with real calls
  * _generate_plan() uses llm_chat from harnessed_agent for LLM-based planning
  * _execute_tool() calls harnessed_execute_tool for real tool execution
  * _get_memory_context() calls harnessed_get_intelligent_memory_context
  * _safety_check() with configurable strict/moderate/lenient modes
  * _try_recovery() for auto-recovery on common failures
  * All session persistence uses real database operations

- Fix reasoning_console.ui: uitype 'textarea' -> 'text' per bricks spec
- Add .gitignore

The reasoning module now:
1. Calls LLM to analyze user request and generate execution plan
2. Safety-checks the plan before execution
3. Executes tools via harnessed_agent's tool system
4. Recovers from common errors automatically
5. Stores session history in database

.gitignore

@@ -0,0 +1,7 @@
+__pycache__/
+*.pyc
+*.pyo
+py3/
+*.egg-info/
+dist/
+build/

wwwroot/reasoning_console.ui

@@ -27,7 +27,7 @@
                     {
                         "name": "request",
                         "label": "推理请求",
-                        "uitype": "textarea",
+                        "uitype": "text",
                         "required": true,
                         "rows": 4,
                         "placeholder": "请输入您的请求，例如：分析当前项目的安全问题并生成修复方案"