The `.m` subdomain will be removed soon, but the per-domain unique-devices metric heavily relies on it as the `.m` subdomain is a line in the metric.
While we should have used the access-method from the beginning, better late then never.
Plan:
[X] Verify that the split using `access_method` gives similar summed-up values and per-domain values for per-domain unique-devices table
[X] Implement the change (unique-devices per-domain only)
* Add `access_method` column to unique_devices_per_domain iceberg tables
* update unique-devices-per-domain computation accordingly
* remove old non-iceberg tables for both per-domain and per-rpject-family
* Update druid loading job for per-domain data (addition of new column)
* Update cassandra loading job for per-domain data (use new column instead of domain-parsing)
[X] Backfill iceberg tables new column in old data using the `.m` subdomain parsing proxy for `mobile web` (we won't have the `mobile app` values backfilled
[] Backfill druid new column using the iceberg table