feat: implement core/ and platform/ with full test coverage

- platform/interface.py: PlatformClient Protocol + Pydantic models (User,
  MessageResponse, UserSettings) — no explicit session management, Master
  handles container lifecycle
- platform/mock.py: MockPlatformClient with simulated latency, [MOCK]
  responses, is_new correctly True only on first creation
- core/protocol.py: unified dataclasses for all events and responses
  (IncomingMessage/Command/Callback, OutgoingMessage/UI/Notification,
  AuthFlow, ChatContext, SettingsAction, etc.)
- core/store.py: StateStore Protocol + InMemoryStore (tests) + SQLiteStore
  (prod) with JSON serialization
- core/chat.py: ChatManager — chat metadata (C1/C2/C3), not container
  lifecycle (that's the platform's job)
- core/auth.py: AuthManager — start_flow / confirm / is_authenticated
- core/settings.py: SettingsManager — get/apply with store cache
- core/handler.py: EventDispatcher — registry-based routing with keys
  (command name, action name, attachment type, "*" catch-all)
- core/handlers/: register_all() + start/new/message/callback/settings
  handlers; voice slot falls back to stub text until voice_handler added
- conftest.py: sys.path fix so local platform/ shadows stdlib platform
- docs/api-contract.md: rewritten for Lambda Lab 3.0 container model

46 tests passing, 0 warnings.

core/handlers/start.py

@@ -0,0 +1,16 @@
+# core/handlers/start.py
+from __future__ import annotations
+
+from core.protocol import IncomingCommand, OutgoingMessage
+
+
+async def handle_start(event: IncomingCommand, auth_mgr, platform, chat_mgr, settings_mgr) -> list:
+    user = await platform.get_or_create_user(event.user_id, event.platform)
+    await auth_mgr.confirm(event.user_id)
+    name = user.display_name or event.user_id
+    text = (
+        f"Добро пожаловать, {name}! Я агент Lambda. Напишите что-нибудь чтобы начать."
+        if user.is_new
+        else f"С возвращением, {name}!"
+    )
+    return [OutgoingMessage(chat_id=event.chat_id, text=text)]