feat(AgentsManager): AI Chat tab for natural language task orchestration

[Copilot]

Users need a conversational interface to launch AgentsManager tasks via natural language — if no suitable ChatGroup exists, the AI proposes creating one (with agent selection and naming), waits for confirmation, then auto-creates and runs the task.

Backend — AiChatAppService

Three [ApiBind(Post)] endpoints:

• SendMessage — Builds a dynamic system prompt injecting all available AgentTemplate and ChatGroup records, calls AI, parses structured JSON response into one of: Message | SuggestRunTask | SuggestCreateGroup | TaskStarted | Error
• ConfirmCreateAndRunGroup — Calls ChatGroupAppService.SetChatGroup() (existing FunctionRender method, not duplicated), then fires ChatGroupService.RunChatGroupInThread() in a background Task.Run with exception logging
• RunExistingGroup — Same fire-and-forget pattern for an already-existing group

// AI system prompt is fully data-driven — no hardcoded agent/group logic
var systemPrompt = BuildSystemPrompt(agentTemplateList, chatGroupList);
// AI returns JSON: {"action":"create_group","suggestion":{...}} or {"action":"run_existing","groupId":N,...}
var responseDto = ParseAiResponse(aiResult.OutputString, agentTemplates, request.AiModelId);

New DTOs: AiChat_SendMessageRequest, AiChat_SuggestedGroupDto, AiChat_RunExistingGroupRequest, AiChat_SendMessageResponse (with AiChatResponseType enum).

AiChatAppService registered in Register.cs.

Frontend — new "AI 对话" tab (fourth)

• Model selector (Chat-type models only), clear button
• Message list: user/AI bubbles rendered via existing task-html-renderer, loading animation
• Inline confirm/cancel buttons on SuggestCreateGroup and SuggestRunTask messages — hidden after action
• Textarea input with Ctrl+Enter shortcut

Key JS methods added to index.js:

• sendAiChatMessage() — posts message + pruned history, handles all response types
• confirmCreateAndRunGroup() / confirmRunExistingGroup() — POST to backend, show success via $message, mark suggestion confirmed
• loadAiChatModelList() — lazy-loads on tab click via handleTabsClick

CSS chat bubble styles appended to index.css.

Original prompt

AI Chat 需要实现效果：让用户用语言对话启动一个AgentsManager的任务。如果没有合适的组做这个任务，则提示用户是否创建一个新的组（并且说明将会参与的Agent以及组的命名等），用户在对话过程中确认后自动创建，并且自动运行任务（类似Prompt优化一样自动触发运行，尽量复用，以及使用EventBus），这个过程需要使用Prompt+functionrender对应的方法来调用实现，不要硬编码。

Created from VS Code.