Follow-up from "Maybe fix the HDF5 multihreading bug"

The following discussion from !336 (merged) should be addressed:

• @npond started a discussion: (+1 comment)

  So, this will work for now, but I'm not sure if its fully thread safe yet, as we open the single file here. I'm not sure how the prediction writer handles multiple workers. I think now this will mean the inputs/outputs are the same for a given jet (as this is handled by the dataset) but that there's a non 0 chance that the additional vars we write to the file are messed up by this.

  I think we should add this as an issue, rather than rolling it in here, as I think the overall effect and consequences are lower from this in the prediction writer.