Add fk between xcom and task instance

[uranusjr]

As discussed previously.

Also improved the dag_run relationship to use the dag_run_id field directly because why not?

[github-actions]

The PR most likely needs to run full matrix of tests because it modifies parts of the core of Airflow. However, committers might decide to merge it quickly and take the risk. If they don't merge it quickly - please rebase it to the latest main at your convenience, or amend the last commit of the PR, and push it with --force-with-lease.

[uranusjr]

Forgot to mention, we can do this directly because the prior select in the migration already filtered out all xcom rows without a matching dag_run row, and since task_instance already has a foreign key to dag_run, it is not possible to have dangling xcom rows when we add the fk constraint in the migration.

[uranusjr]

Eh, need to fix tests now though, we are currently doing a lot of dangling inserts.

[uranusjr]

Alright hopefully this fixes everything. Mostly changing the test code, but I also needed to change how airflow task test works conceptually to make this foreign key work (see the commit referencing this command). Currently airflow task test would create a “fake” DAG run in-memory to run just that task, but this does not work well with database-level foreign key. So instead I make this command always create a “real” DAG run in the database, and delete it (with everything the run generates because ON DELETE CASCADE) after the task is run. Personally I think this is better because now this command is no longer a weird outlier on database integrity, but not sure if this is an acceptable semantic change.

[ashb]

Some good to me