Background events starting over for each sub-file, creating event reuse far beyond expectations

Hi -

I am just reporting an issue found with the background files, which appears to be a consequence of the merging procedure. This was noted by myself and Aidan at ORNL and discussed in the LFHCAL meeting yesterday (June 24).

### Environment: (where does this bug occur, have you tried other environments)
- Which branch (often `main` for latest released):

I am using standard background samples, see the path below

- Which version (or `HEAD` for the most recent on git):

None - just the content of the ROOT files, produced with 26.02.0/craterlake, but the issue is with the merging implementation.

- Any specific OS or system where the issue occurs?

No

- Any special versions of ROOT or Geant4?

No

### Steps to reproduce: (give a step by step account of how to trigger the bug)
1. auto f1 = TFile::Open("root://dtn-eic.jlab.org:1094//volatile/eic/EPIC/FULL/26.02.0/epic_craterlake/Bkg_Exact1S_2us/GoldCt/10um/DIS/NC/10x100/minQ2=1/pythia8NCDIS_10x100_minQ2=1_beamEffects_xAngle=-0.025_hiDiv_1.0001.edm4hep.root")

2. events->Scan("MCParticles.generatorStatus:MCParticles.PDG:MCParticles.momentum.x:MCParticles.momentum.y:MCParticles.momentum.z","MCParticles.generatorStatus>=6000&&MCParticles.generatorStatus<7000")

3. Repeat for different subfiles, e.g. .0002.edm4hep.root, .0003.edm4hep.root, etc. - you will observe the identical particle list being repeated, in order for each file, rather than for a larger set of files.

### Expected Result: (what do you expect when you execute the steps above)

I would expect to see different background events in different files, if you are cycling through a larger set.

### Actual Result: (what do you get when you execute the steps above)

You will see that the background events for each subprocess are repeated in *each* subfile, instead of cycling through what (presumably) is a larger sample of events. Given that some background rates are quite low, e.g. proton beam-gas is 1-2% of the events, and there are 200 events in each file, then the same 2-3 events are repeated many times in the 1000 sub-files of the background sample. This is making background samples which are highly biased since only a few events are actually seen, making "hot spots".