Fix duplicated system stats if you run multiple ETLs in parallel

[jankatins]

If one is running two or more ETLs at the same time and these runs produce a system statistics at the exact same millisecond, we would fail one of these runs with a failure to add the statistic to the DB.

Now:

• A new run_id column which contains the run_id of the run which produces the system statistic. Old rows are migrated to -1 automatically. The default retention period is 30 days, so after 30 days these should be gone.
• We have a compound PK (run_id, timestamp), but unfortunately upgraded DB do not pick this up automatically.
• For that reason we now simply ignore bad writes for system statistics events -> if the mara DB is upgraded without the compound PK, this will catch these cases.
• The UI now only shows the statistics for the specific run (with a fallback for any -1 run_ids in the stats table.

Replaces https://github.com/mara/data-integration/pull/29 Closes: https://github.com/mara/data-integration/issues/22 #29

[martin-loetzsch]

It works:

make migrate-mara-db 
migrate-mara-db: FLASK_APP=app/app.py .venv/bin/flask mara_db.migrate
migrate-mara-db: ALTER TABLE data_integration_system_statistics ADD COLUMN run_id INTEGER DEFAULT -1 NOT NULL;
migrate-mara-db: 
migrate-mara-db: 2 seconds