[ATLASRECTS-6163] Fix ID algorithm scheduling issues in merge steps
This MR addresses two issues with the latest ID reco setup:
- We no longer unintentionally schedule Calo ROI selection or TTVA decoration in merge steps
- We no longer write transient decorations into output ESDs.
This does not fully fix the crashes we see in POOLMergeAthenaMPDESDMMCP0 - there seems to be another conditions issue unrelated to the ones above. Will follow up in a separate MR.