Improve logging about stream parsing errors

During the 22/10/2025 ACADA-LST tests, the data stream contained events with 1848 pixels and 36 samples, but RECO was configured (base structure) for 1855 and 40 samples. As a results the events were silently rejected by the analysis (nbBadMessage not increasing, no logs). We must improve the logging to report this issue explicitely.