This reverts merge request !100455

If a CI migration leaves the database behind the current database
schema, we need to ensure that the schema is brought up-to-date after
the migrations run to prevent other tests from starting from an
inconsistent state.

This commit consolidates the database cleanup for multiple databases
by ensuring:

1. After a migration context is complete, either recreate the
databases (if the number of columns approaches the max) or migrate all
the schemas back to the latest and drop all data from both databases.

2. After a migration example runs, delete any leftover data for that
specific database.  If a CI migration leaves the database behind the
current database schema, we need to ensure that the schema is brought
up-to-date after the migrations run to prevent putting other tests in
an inconsistent state.

Related MR:
https://gitlab.com/gitlab-org/gitlab/-/merge_requests/100233

Closes https://gitlab.com/gitlab-org/gitlab/-/issues/377796

Changelog: fixed

Some GitLab classes, such as `Gitlab::Database::AsWithMaterialized`,
need to check the PostgreSQL version to determine whether certain
keywords can be used. We have some specs that specify exact query
types or check for N+1 queries that might be thrown off by an extra
`SELECT VERSION()` call. To ensure this doesn't happen, retrieve the
version at the beginning of the test.

Note that the initializer `config/initializers/1_postgresql_only.rb`
already attempts to check the PostgreSQL version, but the result isn't
memoized properly for callers that use
`ApplicationRecord.database.version`.

This came up in
https://gitlab.com/gitlab-org/gitlab/-/merge_requests/99097#note_1118427731.

Changelog: changed

Currently the `:migration` helper will run all
migrations in context of `:main`. In some cases
when using `restrict_gitlab_migration gitlab_schema: :gitlab_main`
such migrations will be skipped.

Change `mock_ci_connection` to only setup a mocked
`Ci::ApplicationRecord` connection when running in a single database
mode. When running with multiple databases, we can use the actual
`Ci::ApplicationRecord` connection.

Since LB is initialised globally in a non-decomposed
run we will receive connection of `ActiveRecord::Base`.

Ensure that `GitlabSchemasMetrics` can use FF

This analyzers output is Prometheus metrics observing connections
vs gitlab schemas of executed queries.

It will present well in a single metric how many connections
do cross-join or are misplaced once additional databases
are configured.

The `db_config_name` describes a type of connection being
used by a given query.

Changelog: added