Split-up lcio collection patching and move it to where it is needed

As we are increasingly adding more Gaudi-native algorithms to the steering, there can also be conversions basically anywhere in the future. Therefore, instead of fixing up the lcio event just before the final conversion to edm4hep like now, we should patch the lcio event directly after the offending processors, such that the lcio event is always in a consistent state.

Locate the current lcio event patching immediately before the final conversion to edm4hep, then identify the processors whose output requires patching. Move the patching to follow each offending processor and verify that the lcio event remains consistent wherever a conversion may occur.

Written by the indexing model from the issue text.