Match Qwen reasoning history across effort changes

src/server/request.rs

@@ -556,6 +556,7 @@ pub(super) fn parse_chat_request(
         &request.tools,
         &request.tool_schemas,
         tools_enabled,
+        model,
         protocol,
     )?;
     generation.system_prompt = system;

src/server/tools.rs

@@ -327,10 +327,12 @@ pub(super) fn render_messages(
     tools: &[Value],
     tool_schemas: &[String],
     tools_enabled: bool,
+    model: ModelChoice,
     protocol: Protocol,
 ) -> Result<(String, Vec<ChatTurn>), (u16, String)> {
     validate_tool_results(state, messages, protocol)?;
-    let preserve_reasoning = tools_enabled
+    let preserve_reasoning = model.is_qwen38()
+        || tools_enabled
         || messages.iter().any(|message| {
             matches!(message.role.as_str(), "tool" | "function") || !message.tool_calls.is_empty()
         });