feat: add multi-model delegation (Phase 0) and context compaction (Phase 1)

Phase 0 — Multi-Model Delegation:
- AgentOrchestrator wraps NativeAgent with delegate() for stateless
  single-turn calls to any model tier (fast/default/complex/local)
- DelegationConfig maps task types (compaction, classification, etc.)
  to model tiers
- Delegation prompts for compaction, memory extraction, classification,
  and tool summarisation
- Per-tier usage tracking for cost visibility
- Config schema: agents.delegation and agents.primary_tier

Phase 1 — Context Compaction:
- Token estimation (char/4 heuristic) with context window lookup
- shouldCompact() threshold check against context window percentage
- compactHistory() splits old/recent messages, delegates summary to
  fast tier, returns CompactionResult
- Automatic compaction in AgentOrchestrator.process() when configured
- Force-compact via orchestrator.compact() with session persistence
- Session.replaceHistory() with atomic SQLite transaction
- /compact TUI command with feedback on compacted token counts
- Config schema: compaction.enabled, threshold_pct, keep_turns,
  summary_max_tokens

Tests: 385 passing across 50 files (22 new tests in 2 new test files)

src/daemon/index.ts

@@ -2,7 +2,7 @@ import { Lifecycle } from './lifecycle.js';
 import type { Config, ModelConfig } from '../config/index.js';
 import { AnthropicClient, OpenAIClient, OllamaClient, LlamaCppClient, ModelRouter } from '../models/index.js';
 import type { ModelClient } from '../models/index.js';
-import { NativeAgent } from '../backends/index.js';
+import { AgentOrchestrator, type DelegationConfig } from '../backends/index.js';
 import { SessionStore, SessionManager } from '../session/index.js';
 import { HookEngine } from '../hooks/index.js';
 import { ToolRegistry, ToolExecutor, allBuiltinTools } from '../tools/index.js';
@@ -134,7 +134,8 @@ function createModelRouter(config: Config): ModelRouter {
 
 /**
  * Create the unified message handler for the channel registry.
- * Each channel+sender pair gets its own NativeAgent backed by a persistent session.
+ * Each channel+sender pair gets its own AgentOrchestrator backed by a persistent session.
+ * The orchestrator wraps a NativeAgent and adds delegation to different model tiers.
  */
 function createMessageRouter(deps: {
   sessionManager: SessionManager;
@@ -142,21 +143,39 @@ function createMessageRouter(deps: {
   systemPrompt: string;
   toolRegistry: ToolRegistry;
   toolExecutor: ToolExecutor;
+  config: Config;
 }) {
   // Cache agents by session ID to avoid recreating on every message
-  const agents = new Map<string, NativeAgent>();
+  const agents = new Map<string, AgentOrchestrator>();
 
-  function getOrCreateAgent(channel: string, senderId: string): NativeAgent {
+  function getOrCreateAgent(channel: string, senderId: string): AgentOrchestrator {
     const sessionId = `${channel}:${senderId}`;
     let agent = agents.get(sessionId);
     if (!agent) {
       const session = deps.sessionManager.getSession(channel, senderId);
-      agent = new NativeAgent({
-        modelClient: deps.modelRouter,
+      const delegationConfig: DelegationConfig = {
+        compaction: deps.config.agents.delegation.compaction ?? 'fast',
+        memory_extraction: deps.config.agents.delegation.memory_extraction ?? 'fast',
+        classification: deps.config.agents.delegation.classification ?? 'fast',
+        tool_summarisation: deps.config.agents.delegation.tool_summarisation ?? 'fast',
+        complex_reasoning: deps.config.agents.delegation.complex_reasoning ?? 'complex',
+      };
+      agent = new AgentOrchestrator({
+        modelRouter: deps.modelRouter,
         systemPrompt: deps.systemPrompt,
         session,
         toolRegistry: deps.toolRegistry,
         toolExecutor: deps.toolExecutor,
+        primaryTier: deps.config.agents.primary_tier ?? 'default',
+        delegation: delegationConfig,
+        maxDelegationDepth: deps.config.agents.max_delegation_depth ?? 3,
+        compaction: deps.config.compaction.enabled ? {
+          thresholdPct: deps.config.compaction.threshold_pct,
+          keepTurns: deps.config.compaction.keep_turns,
+          summaryMaxTokens: deps.config.compaction.summary_max_tokens,
+        } : undefined,
+        modelName: deps.config.models.default.model,
+        contextWindow: deps.config.models.default.context_window,
       });
       agents.set(sessionId, agent);
     }
@@ -167,9 +186,26 @@ function createMessageRouter(deps: {
     const agent = getOrCreateAgent(msg.channel, msg.senderId);
 
     // Handle special commands
-    if (msg.metadata?.isCommand && msg.metadata.command === 'reset') {
-      agent.reset();
-      return;
+    if (msg.metadata?.isCommand) {
+      if (msg.metadata.command === 'reset') {
+        agent.reset();
+        return;
+      }
+      if (msg.metadata.command === 'compact') {
+        const result = await agent.compact();
+        if (result && result.compactedCount > 0) {
+          await reply({
+            text: `Compacted ${result.compactedCount} messages: ${result.tokensBefore} → ${result.tokensAfter} tokens`,
+            replyTo: msg.id,
+          });
+        } else {
+          await reply({
+            text: 'Nothing to compact.',
+            replyTo: msg.id,
+          });
+        }
+        return;
+      }
     }
 
     try {
@@ -284,6 +320,7 @@ export async function startDaemon(config: Config): Promise<DaemonContext> {
     systemPrompt,
     toolRegistry,
     toolExecutor,
+    config,
   }));
 
   // Register Telegram adapter