feat: JLPT levels for Japanese, custom practice scenarios, UI improvements

- Add separate level systems: CEFR (A1-C2) for European languages, JLPT (N5-N1) for Japanese
- Store levels per language in new `levels_by_language` JSON field
- Add custom scenario option in AI practice mode
- Show action buttons after practice ends (new dialogue, tasks, words)
- Fix level display across all handlers to use correct level system
- Add Alembic migration for levels_by_language field
- Update all locale files (ru, en, ja) with new keys

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

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

bot/handlers/start.py

@@ -13,6 +13,7 @@ from aiogram.fsm.context import FSMContext
 from database.db import async_session_maker
 from services.user_service import UserService
 from utils.i18n import t
+from utils.levels import get_user_level_for_language
 
 router = Router()
 
@@ -185,10 +186,11 @@ async def pick_theme_from_menu(callback: CallbackQuery, state: FSMContext):
             return
 
         lang = (user.language_interface or 'ru')
+        current_level = get_user_level_for_language(user)
         generating = await callback.message.answer(t(lang, 'words.generating', theme=theme))
         words = await ai_service.generate_thematic_words(
             theme=theme,
-            level=user.level.value,
+            level=current_level,
             count=10,
             learning_lang=user.learning_language,
             translation_lang=user.language_interface,
@@ -200,7 +202,7 @@ async def pick_theme_from_menu(callback: CallbackQuery, state: FSMContext):
             return
 
         # Сохраняем в состояние как в /words
-        await state.update_data(theme=theme, words=words, user_id=user.id, level=user.level.name)
+        await state.update_data(theme=theme, words=words, user_id=user.id, level=current_level)
         await state.set_state(WordsStates.viewing_words)
 
         await show_words_list(callback.message, words, theme)