Concurrent POST /api/v2/backfills causes HTTP 500 + partial data with SQLite metadata DB

### Under which category would you file this issue? Airflow Core ### Apache Airflow version 3.2.2 ### What happened and how to reproduce it? Apologies for AI slop but the issue is clearly reproducible. `<🤖> ` --- Sending multiple concurrent `POST /api/v2/backfills` requests for the same DAG results in HTTP 500 with empty response body, but those failed requests partially commit `backfill_dag_run` rows to SQLite. The partially-created backfill then blocks future creation with "already running backfill". Root cause visible in API server logs: sqlite3.OperationalError: database is locked sqlalchemy.exc.OperationalError: (sqlite3.OperationalError) database is locked `_create_runs_non_partitioned` inserts rows in a loop using `begin_nested()` savepoints. When SQLite's busy timeout fires mid-loop the exception propagates as HTTP 500, but rows flushed by earlier savepoints survive — leaving a partially-initialised backfill. My mise.toml ```tomls [tools] python = "3.12" [env] _.path = [".venv/bin"] AIRFLOW_HOME = "{{config_root}}" AIRFLOW__API__PORT = "48001" AIRFLOW__CORE__EXECUTION_API_SERVER_URL = "http://localhost:48001/execution/" AIRFLOW__CORE__LOAD_EXAMPLES = "false" AIRFLOW__DAG_PROCESSOR__REFRESH_INTERVAL = "1" AIRFLOW__LOGGING__WORKER_LOG_SERVER_HOST = "127.0.0.1" AIRFLOW__SCHEDULER__MAX_DAGRUNS_TO_CREATE_PER_LOOP = "15" AIRFLOW__CORE__MAX_ACTIVE_RUNS_PER_DAG = "15" AIRFLOW__CORE__PARALLELISM = "50" [tasks.setup] description = "Install Airflow into venv" run = """ pip install "apache-airflow==3.2.2" \ --constraint "https://raw.githubusercontent.com/apache/airflow/constraints-3.2.2/constraints-no-providers-3.12.txt" """ ``` **To reproduce:** 1. Create a minimal DAG (`test_backfill_bug.py`): from airflow.sdk import dag, task from datetime import datetime @dag( dag_id="test_backfill_bug", schedule="@daily", start_date=datetime(2020, 1, 1), end_date=datetime(2022, 12, 31), …