If I want to combine a historical with an RCP experiment family from the same model should they have to share the same ensemble ?

Not sure I understand the question. Could you try to rephrase ? usually a RCP run always starts from the end of an historical run. So they go together.

Some ESGF models have different ensembles for the historical experiment and the RCP and I am wondering if I can combine them together as a time serie i.e 1850-2100

I would not do that. I would keep the same ensemble number for hist and scenario part in order to avoid a break n 2005-2006 in the model run.