fix: properly handle thinking mode switching in qwenThinkingMiddleware

Fix issue #11612 where Ollama couldn't disable Qwen3 thinking mode.
The middleware now properly removes existing /think or /no_think suffixes
before adding the correct suffix based on current reasoning_effort setting.

Previously, messages with existing suffixes were skipped, causing thinking
mode to persist even when user switched back to non-thinking mode.

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

Co-Authored-By: Claude <noreply@anthropic.com>

src/renderer/src/aiCore/middleware/qwenThinkingMiddleware.ts

@@ -23,8 +23,10 @@ export function qwenThinkingMiddleware(enableThinking: boolean): LanguageModelMi
             // Process content array
             if (Array.isArray(message.content)) {
               for (const part of message.content) {
-                if (part.type === 'text' && !part.text.endsWith('/think') && !part.text.endsWith('/no_think')) {
-                  part.text += suffix
+                if (part.type === 'text') {
+                  // Remove any existing thinking suffixes first, then add the correct one
+                  const cleanText = part.text.replace(/\s*\/think\s*$/, '').replace(/\s*\/no_think\s*$/, '')
+                  part.text = cleanText + suffix
                 }
               }
             }