refactor: data flow inversion, IPC contract fixes, and frontend adaptation

Data flow inversion:
- Add getHistoryForAgent() in conversations.ts — Agent now reads
  history directly from SQLite instead of receiving it from frontend
- Remove historyMessages parameter from agent:runStream IPC chain,
  add maxHistoryRounds for configurable context window
- Frontend (useAIChat.ts) pre-creates conversation before Agent call,
  removes history message collection logic

IPC fixes:
- Restore llm:chat and llm:chatStream handlers using pi-ai's
  completeSimple/streamSimple (P0-1: SQLLab compatibility)
- Send agent:complete with aborted flag in all abort paths to prevent
  hanging Promises in renderer (P0-2)
- Fix createConversation parameter order to match preload/data layer
  contract: (sessionId, title?) instead of (title, sessionId?) (P1)

Preload/frontend:
- Update preload API signatures and type declarations
- Use shared types for TokenUsage and AgentRuntimeStatus

electron/preload/index.d.ts

@@ -1,5 +1,6 @@
 import { ElectronAPI } from '@electron-toolkit/preload'
 import type { AnalysisSession, MessageType, ImportProgress, ExportProgress } from '../../src/types/base'
+import type { TokenUsage, AgentRuntimeStatus } from '../shared/types'
 import type {
   MemberActivity,
   MemberNameHistory,
@@ -582,29 +583,7 @@ interface RAGConfig {
   topK?: number
 }
 
-// Token 使用量类型
-interface TokenUsage {
-  promptTokens: number
-  completionTokens: number
-  totalTokens: number
-}
-
-interface AgentRuntimeStatus {
-  phase: 'preparing' | 'thinking' | 'tool_running' | 'responding' | 'completed' | 'aborted' | 'error'
-  round: number
-  toolsUsed: number
-  currentTool?: string
-  contextTokens: number
-  contextWindow: number
-  contextUsage: number
-  totalUsage: TokenUsage
-  nodeCount?: number
-  tagCount?: number
-  segmentSize?: number
-  checkoutCount?: number
-  activeAnchorNodeId?: string | null
-  updatedAt: number
-}
+// TokenUsage & AgentRuntimeStatus — imported from electron/shared/types.ts
 
 // Agent 相关类型
 interface AgentStreamChunk {
@@ -659,10 +638,10 @@ interface AgentApi {
     userMessage: string,
     context: ToolContext,
     onChunk?: (chunk: AgentStreamChunk) => void,
-    historyMessages?: Array<{ role: 'user' | 'assistant'; content: string }>,
     chatType?: 'group' | 'private',
     promptConfig?: PromptConfig,
-    locale?: string
+    locale?: string,
+    maxHistoryRounds?: number
   ) => { requestId: string; promise: Promise<{ success: boolean; result?: AgentResult; error?: string }> }
   abort: (requestId: string) => Promise<{ success: boolean; error?: string }>
 }