feat: implement full context-scoped AI chat UI in AIChatPanel

- Added AIChatPanel component with context header, user and AI message handling.
- Integrated streaming responses via IPC and error handling for chat mutations.
- Enhanced user experience with input handling and auto-scrolling features.
- Updated AppShell to derive AI chat context from the current route.
- Introduced ScrollArea component for better scrolling behavior in various dialogs.
- Added support for Tailwind typography and improved global styles.
- Updated project and task dialogs to utilize ScrollArea for better UX.

src/main/ai/orchestrator.ts

@@ -6,7 +6,6 @@
  */
 import { Annotation, StateGraph, START, END } from '@langchain/langgraph';
 import { SystemMessage, HumanMessage, type BaseMessage } from '@langchain/core/messages';
-import { z } from 'zod';
 import { eq, asc } from 'drizzle-orm';
 import { getDb } from '../db';
 import { projects, tasks, checkpoints, notes, clients } from '../db/schema';
@@ -178,12 +177,6 @@ If the user asks about specific note contents that aren't included here, let the
 // LangGraph State
 // ---------------------------------------------------------------------------
 
-const RouteSchema = z.object({
-  route: z.enum(['project', 'knowledge', 'general']).describe(
-    'Which specialist agent should handle this request',
-  ),
-});
-
 const OrchestratorState = Annotation.Root({
   /** The user's original message */
   userMessage: Annotation<string>(),
@@ -207,25 +200,29 @@ type State = typeof OrchestratorState.State;
 // Graph nodes
 // ---------------------------------------------------------------------------
 
-/** Node 1: Classify intent using structured output */
+/** Node 1: Classify intent using plain-text extraction (works with all providers) */
 async function classifyIntent(state: State): Promise<Partial<State>> {
   const llm = await getLLM();
   if (!llm) throw new Error('AI provider not configured. Please add your token in Settings.');
 
-  const routerLLM = llm.withStructuredOutput(RouteSchema);
-
-  const result = await routerLLM.invoke([
+  const response = await llm.invoke([
     new SystemMessage(
       `You are a routing classifier for Adiuva, a project management workspace.
-Classify the user's message into one of these categories:
-- "project": Question about a specific project (tasks, notes, checkpoints, progress, summaries)
-- "knowledge": Cross-project or historical question (e.g., "what did we decide about X?", "find notes about Y")
-- "general": Everything else (general help, scheduling, task overviews, workspace summaries)`,
+Classify the user's message into exactly one category. Reply with ONLY the category name, nothing else.
+
+Categories:
+- project: Question about a specific project (tasks, notes, checkpoints, progress, summaries)
+- knowledge: Cross-project or historical question (e.g., "what did we decide about X?", "find notes about Y")
+- general: Everything else (general help, scheduling, task overviews, workspace summaries)`,
     ),
     new HumanMessage(state.userMessage),
   ]);
 
-  return { route: result.route };
+  const text = (typeof response.content === 'string' ? response.content : '').trim().toLowerCase();
+  const validRoutes = ['project', 'knowledge', 'general'] as const;
+  const route = validRoutes.find((r) => text.includes(r)) ?? 'general';
+
+  return { route };
 }
 
 /** Node 2a: Project agent — answer project-scoped questions */