Allow null start_date in DagRun Pydantic model to fix rerun regression (#61381)

[v4xsh]

Fix Pydantic ValidationError when rerunning cleared DagRun

Closes #61381

---

Problem

When rerunning or clearing DAG runs (especially those created in older Airflow versions), the scheduler may raise:

ValidationError: dag_run.start_date
Input should be a valid datetime

This occurs because the Pydantic DagRun model required start_date to be a non-null datetime, while cleared or queued DAG runs can legitimately have start_date=None in the database.

---

Root Cause

The Pydantic DagRun model in:

airflow-core/src/airflow/api_fastapi/execution_api/datamodels/taskinstance.py

did not allow start_date=None, causing a ValidationError when constructing DagRunContext during scheduling.

---

Changes Made

1️⃣ Allow start_date to be nullable

- start_date: UtcDateTime
+ start_date: UtcDateTime | None = None

This aligns the Pydantic model with:

• ORM definition
• Database state
• Scheduler behavior

2️⃣ Add default for partition_key

- partition_key: str | None
+ partition_key: str | None = None

Ensures consistent optional behavior.

---

Tests Added

New test file:

airflow-core/tests/unit/api_fastapi/execution_api/datamodels/test_taskinstance.py

Includes:

• test_dagrun_model_accepts_null_start_date
• test_dagrun_context_accepts_null_start_date

These verify:

• The Pydantic DagRun model accepts start_date=None
• DagRunContext can be constructed without raising ValidationError

---

Result

• Scheduler no longer crashes when rerunning cleared DAG runs
• Backward compatibility with existing DAG runs is restored
• Model validation now reflects actual database behavior

---