test(migrations): Scope reverse idempotency snapshot to Snuba databases (#8592)

`test_reverse_idempotency_all` snapshots `system.tables` before and
after reversing each group's latest migration twice. The snapshot
included ClickHouse's own `system.*` tables, and ClickHouse creates log
tables such as `system.query_metric_log` lazily in the background. When
one of those appeared mid-test, the before and after snapshots differed
and the test failed for reasons unrelated to migrations.

That happened on [attempt 1 of this
run](https://github.com/getsentry/snuba/actions/runs/39140651685/job/117714460279):
the failure diff contains `system.query_metric_log`, and the same image
passed on retry. Because the whole workflow was rerun, 34 already-green
jobs ran again, which cost about 184 runner-minutes.

The snapshot now only covers the databases of Snuba's configured
clusters, the same set `_drop_all_tables` uses. It selects `(database,
name, create_table_query)` ordered by `(database, name)`. Every group,
the double reversal, and the comparison of all Snuba tables across
groups are unchanged.

A side benefit: the failure output used to print every `system` table
definition (about 15k log lines). It now only prints Snuba tables.

<!-- junior-request-attribution:start -->
via **Joshua Li**.
<!-- junior-request-attribution:end -->

<!-- junior-session-footer:start -->
<!-- junior-conversation-id:local%3Aweb%3Ad2c357b94d471741d23c9928 -->

--

[View Junior Session in
Sentry](https://sentry.sentry.io/explore/conversations/local%3Aweb%3Ad2c357b94d471741d23c9928/?project=4510944073809921)

<!-- junior-session-footer:end -->

Co-authored-by: sentry-junior[bot] <264270552+sentry-junior[bot]@users.noreply.github.com>
Co-authored-by: Joshua Li <joshua.li@sentry.io>

Author: sentry-junior[bot]

tests/migrations/test_runner.py

@@ -318,6 +318,15 @@ def test_reverse_idempotency_all() -> None:
     """
     runner = Runner()
     groups = get_active_migration_groups()
+    # Snapshot only Snuba's databases: ClickHouse creates `system.*` log tables
+    # (e.g. `system.query_metric_log`) lazily in the background, which made the
+    # before/after comparison flaky. `system.tables` has no guaranteed row
+    # order, so order by table identity.
+    snuba_databases = ", ".join(sorted({f"'{cluster.get_database()}'" for cluster in CLUSTERS}))
+    snapshot_query = (
+        "SELECT database, name, create_table_query FROM system.tables "
+        f"WHERE database IN ({snuba_databases}) ORDER BY database, name"
+    )
     group_migrations = {
         group.value: runner._get_pending_migrations_for_group(group) for group in groups
     }
@@ -349,14 +358,6 @@ def reverse_twice(migration: MigrationKey = migration) -> None:
                     ClickhouseClientSettings.MIGRATE
                 )
 
-                # `system.tables` has no guaranteed row order, and the
-                # reverse_twice() cycle in between can perturb the order rows
-                # are returned in (e.g. ON CLUSTER DDL re-registering tables).
-                # Order the snapshots so the comparison reflects the set of
-                # tables and their definitions, not the iteration order.
-                snapshot_query = (
-                    "SELECT create_table_query FROM system.tables ORDER BY create_table_query"
-                )
                 before_state = cluster_connection.execute(snapshot_query).results
                 reverse_twice()
                 after_state = cluster_connection.execute(snapshot_query).results