feat: add AI Training Mode with Bedrock exercise generation

[kiro-agent]

This pull request was generated by @kiro-agent 👻

Comment with /kiro fix to address specific feedback or /kiro all to address everything.
_{Learn about Kiro autonomous agent}

---

Summary

Implements the full AI Training Mode feature as specified in .kiro/specs/ai-training-mode/. This adds a new AI_TRAINING mode that uses Amazon Bedrock to generate contextual language exercises (fill-in-the-blank, verb conjugation, preposition selection, sentence completion) based on selected vocabulary words.

Changes

Backend - Domain Model & GraphQL Schema

• Extended TrainingMode type with 'AI_TRAINING' value
• Added AIExercise interface (prompt, options, correctOptionIndex, exerciseType, sourceWord)
• Added aiExercises?: AIExercise[] to TrainingExecution
• Updated GraphQL schema with AI_TRAINING enum value, AIExercise type, and aiExercises field on TrainingExecution

Backend - AI Service (ai-service.ts)

• Added generateExercises() method: builds structured Bedrock prompt, calls InvokeModelCommand, parses and validates response
• Added parseAndValidateExercises() with graceful degradation (filters out invalid exercises, logs warnings)
• Respects existing rate limiting and usage logging

Backend - Training Service (training-service.ts)

• createTraining: accepts AI_TRAINING mode (static and randomized)
• startTraining: new AI_TRAINING branch - selects words, fetches full word details + language info, calls AIService.generateExercises, stores aiExercises on execution
• submitAnswer: new AI_TRAINING branch - compares selected option index against correctOptionIndex, tracks completion

Frontend - Flutter

• Updated training_provider.dart to fetch aiExercises fields in GraphQL mutations
• Added AI Training ChoiceChip in training_creation_screen.dart
• Created ai_exercise_widget.dart - displays exercise type, prompt, answer options as tappable buttons, correct/incorrect feedback
• Updated training_execution_screen.dart to handle AI_TRAINING mode with progress tracking

Tests

• ai-service.test.ts - 11 unit tests covering exercise generation, rate limiting, error handling, parsing
• ai-service.property.test.ts - Property tests for prompt construction, parsing/validation, round-trip serialization
• ai-training-service.test.ts - 7 unit tests for AI training creation, start, submission, completion
• ai-training-service.property.test.ts - 4 property tests for creation, start, submission, backward compatibility

All 28 tasks completed

All tasks from .kiro/specs/ai-training-mode/tasks.md have been implemented and marked as complete.

Notes

• No new DynamoDB tables, GSIs, or CDK stack changes required
• Lambda resolvers require no changes (they delegate to TrainingService)
• Existing TEXT_INPUT and MULTIPLE_CHOICE trainings are unaffected (backward compatible)
• Tests could not be executed in the sandbox environment (no npm/node_modules available) - please run npm test locally to verify