feat: support multiple event proposals in single AI response

- Change proposedChange to proposedChanges array in ChatMessage type
- Add unique id and individual respondedAction to each ProposedEventChange
- Implement arrow navigation UI for multiple proposals with "Event X von Y" counter
- Add updateProposalResponse() method for per-proposal confirm/reject tracking
- GPTAdapter now collects multiple tool call results into proposals array
- Add RRULE documentation to system prompt (separate events for different times)
- Fix RRULE parsing to strip RRULE: prefix if present
- Add log summarization for large args (conversationHistory, existingEvents)
- Keep proposedChanges logged in full for debugging AI issues

CLAUDE.md

@@ -205,15 +205,17 @@ src/
 - `User`: id, email, userName, passwordHash?, createdAt?, updatedAt?
 - `CalendarEvent`: id, userId, title, description?, startTime, endTime, note?, isRecurring?, recurrenceRule?
 - `ExpandedEvent`: Extends CalendarEvent with occurrenceStart, occurrenceEnd (for recurring event instances)
-- `ChatMessage`: id, conversationId, sender ('user' | 'assistant'), content, proposedChange?, respondedAction?
-- `ProposedEventChange`: action ('create' | 'update' | 'delete'), eventId?, event?, updates?
+- `ChatMessage`: id, conversationId, sender ('user' | 'assistant'), content, proposedChanges?
+- `ProposedEventChange`: id, action ('create' | 'update' | 'delete'), eventId?, event?, updates?, respondedAction?
+  - Each proposal has unique `id` (e.g., "proposal-0") for individual confirm/reject
+  - `respondedAction` tracks user response per proposal (not per message)
 - `Conversation`: id, userId, createdAt?, updatedAt? (messages loaded separately via lazy loading)
 - `CreateUserDTO`: email, userName, password (for registration)
 - `LoginDTO`: identifier (email OR userName), password
 - `CreateEventDTO`: Used for creating events AND for AI-proposed events
 - `GetMessagesOptions`: Cursor-based pagination with `before?: string` and `limit?: number`
 - `ConversationSummary`: id, lastMessage?, createdAt? (for conversation list)
-- `UpdateMessageDTO`: respondedAction? (for marking messages as confirmed/rejected)
+- `UpdateMessageDTO`: proposalId?, respondedAction? (for marking individual proposals as confirmed/rejected)
 - `RespondedAction`: 'confirm' | 'reject' (tracks user response to proposed events)
 - `Day`: "Monday" | "Tuesday" | ... | "Sunday"
 - `Month`: "January" | "February" | ... | "December"
@@ -275,6 +277,14 @@ export class GPTAdapter implements AIProvider { ... }
 
 The decorator uses a Proxy to intercept method calls lazily, preserves sync/async nature, and logs start/completion/failure with duration.
 
+**Log Summarization:**
+The `@Logged` decorator automatically summarizes large arguments to keep logs readable:
+- `conversationHistory` → `"[5 messages]"`
+- `existingEvents` → `"[3 events]"`
+- `proposedChanges` → logged in full (for debugging AI issues)
+- Long strings (>100 chars) → truncated
+- Arrays → `"[Array(n)]"`
+
 **Client Logging:**
 - `react-native-logs` with namespaced loggers (apiLogger, storeLogger)
 - ApiClient logs all requests with method, endpoint, status, duration
@@ -339,10 +349,12 @@ NODE_ENV=development        # development = pretty logs, production = JSON
   - `utils/recurrenceExpander`: expandRecurringEvents() using rrule library for RRULE parsing
   - `ChatController`: getConversations(), getConversation() with cursor-based pagination support
   - `ChatService`: getConversations(), getConversation(), processMessage() uses real AI or test responses (via USE_TEST_RESPONSES), confirmEvent()/rejectEvent() update respondedAction and persist response messages
-  - `MongoChatRepository`: Full CRUD implemented (getConversationsByUser, createConversation, getMessages with cursor pagination, createMessage, updateMessage)
-  - `ChatRepository` interface: updateMessage() added for respondedAction tracking
-  - `GPTAdapter`: Full implementation with OpenAI GPT (gpt-5-mini model), function calling for calendar operations
+  - `MongoChatRepository`: Full CRUD implemented (getConversationsByUser, createConversation, getMessages with cursor pagination, createMessage, updateMessage, updateProposalResponse)
+  - `ChatRepository` interface: updateMessage() and updateProposalResponse() for per-proposal respondedAction tracking
+  - `GPTAdapter`: Full implementation with OpenAI GPT (gpt-4o-mini model), function calling for calendar operations, collects multiple proposals per response
   - `ai/utils/`: Provider-agnostic shared utilities (systemPrompt, toolDefinitions, toolExecutor, eventFormatter)
+  - `ai/utils/systemPrompt`: Includes RRULE documentation - AI knows to create separate events when times differ by day
+  - `utils/recurrenceExpander`: Handles RRULE parsing, strips `RRULE:` prefix if present (AI may include it)
   - `logging/`: Structured logging with pino, pino-http middleware, @Logged decorator
   - All repositories and GPTAdapter decorated with @Logged for automatic method logging
   - CORS configured to allow X-User-Id header
@@ -373,6 +385,9 @@ NODE_ENV=development        # development = pretty logs, production = JSON
   - Supports events from adjacent months visible in grid
   - Uses `useFocusEffect` for automatic reload on tab focus
 - Chat screen fully functional with FlashList, message sending, and event confirm/reject
+  - **Multiple event proposals**: AI can propose multiple events in one response
+  - Arrow navigation between proposals with "Event X von Y" counter
+  - Each proposal individually confirmable/rejectable
   - Messages persisted to database via ChatService and loaded on mount
   - Tracks conversationId for message continuity across sessions
   - ChatStore with addMessages() for bulk loading, chatMessageToMessageData() helper