DPP-334 Manually partition the event table

[rautenrieth-da]

This PR removes the automatic partitioning of the participant_events table in favor of a manual partitioning. This is easier to port to other databases, and removes unnecessary fields from individual partitions.

Schema changes

• The schema now contains 4 distinct tables: participant_events_divulgence, participant_events_create, participant_events_consuming_exercise, and participant_events_non_consuming_exercise. The individual tables are leaner than before - they don't contain useless fields, and the remaining fields have been made NOT NULL where possible.
• There is now a view participant_events that contains exactly the same fields as the previous partitioned table of the same name.
• In the indexer, events are grouped by their kind, and inserted directly into the respective tables. This increases the number of SQL statements executed (previously: 1 batch of mixed events, now: 4 batches of events of the same kind).

Testing

The conformance test tool is now run against the append-only schema on Postgres, using ledger-on-sql.

Future work

• Make fields of DBDTOV1.Event* non-optional where applicable.
• Change all queries to select from the individual partitions instead of from the view, where it makes sense.
• Implement the removal of partially written divulgence events

[meiersi-da]

Thanks @rautenrieth-da for this good work! As you said, it is a bit boilerplaty, but it is also simple, which I like. Comments added. Nothing major.

[tudor-da]

LGTM. Tested in an E2E run with no observed performance impact.