fix: replace fmt.Sprintf in logger calls with structured fields and add LoopDelay default

fmt.Sprintf inside zap logger calls defeats structured logging,
making log aggregation and filtering harder. All 6 sites now use
proper zap fields (zap.Uint32, zap.Uint8, zap.String).

LoopDelay had no viper.SetDefault, so omitting it from config.json
caused a zero-value (0 ms) busy-loop in the recv loop. Default is
now 50 ms, matching config.example.json.

server/channelserver/handlers_data.go

@@ -180,7 +180,7 @@ func handleMsgMhfLoaddata(s *Session, p mhfpacket.MHFPacket) {
 
 	data, err := s.server.charRepo.LoadColumn(s.charID, "savedata")
 	if err != nil || len(data) == 0 {
-		s.logger.Warn(fmt.Sprintf("Failed to load savedata (CID: %d)", s.charID), zap.Error(err))
+		s.logger.Warn("Failed to load savedata", zap.Uint32("charID", s.charID), zap.Error(err))
 		_ = s.rawConn.Close() // Terminate the connection
 		return
 	}