feat: implement AI Director & Narrative Voting System (Phase 9)

Add complete AI Director system that transforms the survival simulation
into a user-driven interactive story with audience voting.

Backend:
- Add DirectorService for LLM-powered plot generation with fallback templates
- Add VoteManager for dual-channel voting (Twitch + Unity)
- Integrate 4-phase game loop: Simulation → Narrative → Voting → Resolution
- Add vote command parsing (!1, !2, !A, !B) in Twitch service
- Add type-safe LLM output handling with _coerce_int() helper
- Normalize voter IDs for case-insensitive duplicate prevention

Unity Client:
- Add NarrativeUI for cinematic event cards and voting progress bars
- Add 7 new event types and data models for director/voting events
- Add delayed subscription coroutine for NetworkManager timing
- Sync client timer with server's remaining_seconds to prevent drift

Documentation:
- Update README.md with AI Director features, voting commands, and event types

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

backend/app/twitch_service.py

@@ -74,6 +74,18 @@ class TwitchBot(commands.Bot):
         # Log the message for debugging
         logger.info(f"Twitch chat [{username}]: {content}")
 
+        # Phase 9: Check for vote commands first (!1, !2, !A, !B)
+        vote_index = self._game_engine.parse_vote_command(content)
+        if vote_index is not None:
+            try:
+                voted = await self._game_engine.process_vote(username, vote_index, "twitch")
+                if voted:
+                    logger.info(f"Vote recorded: {username} -> option {vote_index + 1}")
+                return  # Don't process as regular command
+            except Exception as e:
+                logger.error(f"Error processing vote: {e}")
+                return
+
         # Forward to game engine for command processing
         try:
             await self._game_engine.process_command(username, content)