async parsing + tex caching

Plugins/ZShell/Llm/messagemodel.cpp

@@ -2,6 +2,8 @@
 
 #include "session.hpp"
 
+#include <algorithm>
+
 namespace ZShell::llm {
 
 ChatMessageModel::ChatMessageModel(ChatSession* session, QObject* parent)
@@ -98,7 +100,12 @@ void ChatMessageModel::clear() {
 
 void ChatMessageModel::loadMessages(QList<ChatMessage*> messages) {
 	beginResetModel();
-	qDeleteAll(m_messages);
+	// The new list may share rows with the current one (live rows kept
+	// in front of fetched rows); delete only what is truly gone.
+	for (ChatMessage* message : m_messages)
+		if (std::find(messages.begin(), messages.end(), message)
+		    == messages.end())
+			delete message;
 	m_messages = std::move(messages);
 	endResetModel();