Use selective_matching=True for lead-lead config
This significantly reduces the memory usage in the calo reconstruction (notably in FuturePhotonMatchAlg, FutureElectronMatchAlg), which hopefully allows the throughput test to fit in the available memory.
To be confirmed if this makes physics sense.
Related to Rec!2606 (merged) and lhcb-datapkg/PRConfig!201 (merged)