fix: SQLite engine fallback and conversation_id in summaries

- Fall back to query.NewSQLiteEngine when the Parquet cache is
  incomplete so remote TUI endpoints work on fresh installs
- Add conversation_id to list/search SELECT columns and update
  scanMessageRows so summary responses populate ConversationID
  consistently with the detail endpoint

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: wesm

cmd/msgvault/cmd/serve.go

@@ -84,25 +84,26 @@ func runServe(cmd *cobra.Command, args []string) error {
 	}
 
 	// Create query engine for TUI aggregate support.
-	// Requires a complete Parquet cache; without one the engine
-	// would succeed at creation but fail on first query.
+	// Prefer DuckDB over Parquet when the cache is complete; otherwise
+	// fall back to SQLite so remote endpoints still work on fresh installs.
 	analyticsDir := cfg.AnalyticsDir()
-	var engine *query.DuckDBEngine
+	var engine query.Engine
 	if query.HasCompleteParquetData(analyticsDir) {
-		var engineErr error
-		engine, engineErr = query.NewDuckDBEngine(
+		duckEngine, engineErr := query.NewDuckDBEngine(
 			analyticsDir, dbPath, s.DB(),
 		)
 		if engineErr != nil {
-			logger.Warn("query engine not available - aggregate endpoints will return 503",
+			logger.Warn("DuckDB engine failed, falling back to SQLite",
 				"error", engineErr)
+			engine = query.NewSQLiteEngine(s.DB())
+		} else {
+			engine = duckEngine
 		}
 	} else {
-		logger.Warn("parquet cache not built - run 'msgvault build-cache' to enable aggregate endpoints")
-	}
-	if engine != nil {
-		defer engine.Close()
+		logger.Info("parquet cache not built - using SQLite engine (run 'msgvault build-cache' for faster aggregates)")
+		engine = query.NewSQLiteEngine(s.DB())
 	}
+	defer engine.Close()
 
 	// Create OAuth manager
 	oauthMgr, err := oauth.NewManager(cfg.OAuth.ClientSecrets, cfg.TokensDir(), logger)

internal/store/api.go

@@ -44,6 +44,7 @@ func (s *Store) ListMessages(offset, limit int) ([]APIMessage, int64, error) {
 	query := `
 		SELECT
 			m.id,
+			COALESCE(m.conversation_id, 0) as conversation_id,
 			COALESCE(m.subject, '') as subject,
 			COALESCE(p.email_address, '') as from_email,
 			COALESCE(m.sent_at, m.received_at, m.internal_date) as sent_at,
@@ -173,6 +174,7 @@ func (s *Store) SearchMessages(query string, offset, limit int) ([]APIMessage, i
 	ftsQuery := `
 		SELECT
 			m.id,
+			COALESCE(m.conversation_id, 0) as conversation_id,
 			COALESCE(m.subject, '') as subject,
 			COALESCE(p.email_address, '') as from_email,
 			COALESCE(m.sent_at, m.received_at, m.internal_date) as sent_at,
@@ -250,6 +252,7 @@ func (s *Store) searchMessagesLike(query string, offset, limit int) ([]APIMessag
 	searchQuery := `
 		SELECT
 			m.id,
+			COALESCE(m.conversation_id, 0) as conversation_id,
 			COALESCE(m.subject, '') as subject,
 			COALESCE(p.email_address, '') as from_email,
 			COALESCE(m.sent_at, m.received_at, m.internal_date) as sent_at,
@@ -288,15 +291,15 @@ func (s *Store) searchMessagesLike(query string, offset, limit int) ([]APIMessag
 	return messages, total, nil
 }
 
-// scanMessageRows scans the standard 7-column message row set.
+// scanMessageRows scans the standard 8-column message row set.
 // Uses string scanning for dates to handle all SQLite datetime formats robustly.
 func scanMessageRows(rows *sql.Rows) ([]APIMessage, []int64, error) {
 	var messages []APIMessage
 	var ids []int64
 	for rows.Next() {
 		var m APIMessage
 		var sentAtStr sql.NullString
-		err := rows.Scan(&m.ID, &m.Subject, &m.From, &sentAtStr, &m.Snippet, &m.HasAttachments, &m.SizeEstimate)
+		err := rows.Scan(&m.ID, &m.ConversationID, &m.Subject, &m.From, &sentAtStr, &m.Snippet, &m.HasAttachments, &m.SizeEstimate)
 		if err != nil {
 			return nil, nil, err
 		}