async parsing + tex caching

Plugins/ZShell/Llm/session.hpp

@@ -43,10 +43,17 @@ class ChatSession : public QObject {
 	[[nodiscard]] qint64 updatedAtMs() const { return m_updatedAt; }
 	[[nodiscard]] bool pinned() const { return m_pinned; }
 	[[nodiscard]] int messageCount() const { return m_messageCount; }
-	// Loads the messages from the store on first access.
+	// The messages model; the first access starts the (async) load
+	// from the store.
 	[[nodiscard]] ChatMessageModel* messagesModel();
 	void ensureLoaded();
+	// True once the async load from the store has finished.
 	[[nodiscard]] bool isLoaded() const { return m_loaded; }
+	// The model without triggering a load (ChatStore use during the
+	// load itself).
+	[[nodiscard]] ChatMessageModel* model() const { return m_model; }
+	void markLoaded();
+	[[nodiscard]] bool takeClearPending();
 
 	[[nodiscard]] LlmClient* client() const;
 	[[nodiscard]] int lastTokenCount() const { return m_lastTokenCount; }
@@ -80,6 +87,8 @@ class ChatSession : public QObject {
 	void updatedAtChanged();
 	void pinnedChanged();
 	void messageCountChanged();
+	// The messages finished loading from the store.
+	void loaded();
 
 	private:
 	void onModelRowsChanged();
@@ -96,7 +105,9 @@ class ChatSession : public QObject {
 	bool m_pinned = false;
 	int m_messageCount = 0;
 	ChatMessageModel* m_model = nullptr;
+	bool m_loadRequested = false;
 	bool m_loaded = false;
+	bool m_clearPending = false;
 	int m_lastTokenCount = 0;
 };