Event mapper payload consistency

[cmgrote]

On testing the event mapper bugs in #17 a number of different times it seems that the payloads that are generated are not consistent. In particular, sometimes certain objects are identified as having changed (typically around data_file_records and their contained data_file_fields) and other times not. When it did appear, it looked like it was to do with the items array of relationship properties being ordered differently: or at last being detected as being ordered differently.

Needs deeper investigation: have added thorough debug logging to output the diffs that are calculated and the full source and target objects that were diff'd. Next step is to reproduce the problematic diffs (with debug logging on) to see whether the problem appears to be from:

• different items array ordering from IGC from one query to another (within a relationship property, so sorting cannot be applied)
• spurious JsonDiff results despite the compared objects actually having the same array ordering
• something else?

The issue is that the diffs being generated include either replace or remove ops, causing DELETE_PURGED_RELATIONSHIP_EVENTs to be triggered -- and in many cases the last event raised for a given relationship is this DELETE_PURGED_RELATIONSHIP_EVENT which would mean the relationship is entirely removed from the cohort (never replaced with a NEW_RELATIONSHIP_EVENT).

[cmgrote]

Parking as an enhancement, as unlikely the event mapper will be reinstated.