Fix Task SDK dry-run mode crashing on task instance start

[Eason09053360]

Why

dry_run=True is the Task SDK's documented way to run or debug a task locally without an API
server. It has been broken for a long time: noop_handler's hand-written dag_run payload has
not changed since the Task SDK became its own distribution, while the generated DagRun model
gained five required fields (data_interval_start, data_interval_end, end_date, state,
partition_key). task_instances.start() raises a ValidationError on the first call, and the
supervisor then kills the child process it has already forked.

Client(base_url=None, dry_run=True, token="").task_instances.start(uuid7(), 1234, datetime.now(tz=timezone.utc))
# pydantic_core._pydantic_core.ValidationError: 5 validation errors for TIRunContext

What

noop_handler serves a TIRunContext built from the generated models instead of a literal dict,
so the next required field added to DagRun fails mypy-task-sdk at the definition rather than
surfacing at runtime under dry_run.

TestClient::test_dry_run is an existing test this PR rewrites, so it is worth saying why: it
compared the payload against a hand-copied duplicate of itself and fetched it with client.get(),
never reaching TIRunContext.model_validate_json(). That is how the drift shipped — the test
passes on main with the bug live. It now goes through task_instances.start() and fails
without the fix.

Also drops a duplicate datetime import in the TYPE_CHECKING block that ruff flags (TC004)
once datetime is used at runtime.

How to test

uv run --project task-sdk pytest task-sdk/tests/task_sdk/api/test_client.py -k dry_run

---

Was generative AI tooling used to co-author this PR?

• Yes — Claude Code (Opus 5)

Generated-by: Claude Code (Opus 5) following the guidelines

[potiuk]

Hello @Eason09053360 - thank you for your contributions to Apache Airflow!

The Airflow community has introduced a limit of 5 open pull requests at a time for contributors without write access to the repository. You currently have 33 open pull requests, so - as a one-time step of introducing the limit - we closed the ones where maintainers have not engaged yet:

• Return only the latest Dag version from iter_all_latest_version_dags #70935 - Return only the latest Dag version from iter_all_latest_version_dags
• Fix airflow config get-value silently succeeding on a missing option #72145 - Fix airflow config get-value silently succeeding on a missing option
• Fix airflowctl tracebacks on an unusable credentials file #72468 - Fix airflowctl tracebacks on an unusable credentials file
• Make suppress_logs_and_warning CLI decorator return the wrapped result #72600 - Make suppress_logs_and_warning CLI decorator return the wrapped result
• Fix dateless dag.test and tasks test destroying unrelated Dag runs #72656 - Fix dateless dag.test and tasks test destroying unrelated Dag runs
• Fix airflowctl commands crashing instead of printing their result #72675 - Fix airflowctl commands crashing instead of printing their result
• Show a clean error when airflow backfill create targets an unknown Dag #72893 - Show a clean error when airflow backfill create targets an unknown Dag
• Fix tasks clear --task-regex matching as a plain substring #72941 - Fix tasks clear --task-regex matching as a plain substring
• Speed up airflow dags list on deployments with many Dags #73033 - Speed up airflow dags list on deployments with many Dags
• Fix airflow dags next-execution crashing on --field data_interval.start #73070 - Fix airflow dags next-execution crashing on --field data_interval.start
• Fix airflow connections import corrupting JSON values in .env files #73083 - Fix airflow connections import corrupting JSON values in .env files
• Fix structlog log assertions never matching a single-field dict #73112 - Fix structlog log assertions never matching a single-field dict
• Fix misleading type errors from expand_kwargs and default_args #73120 - Fix misleading type errors from expand_kwargs and default_args
• Fix bare traceback from kubernetes cleanup-pods without kube config #73121 - Fix bare traceback from kubernetes cleanup-pods without kube config
• Fix airflowctl pools export/import dropping pool team ownership #73132 - Fix airflowctl pools export/import dropping pool team ownership
• Mask extra and credentials in connections list --hide-sensitive #73157 - Mask extra and credentials in connections list --hide-sensitive
• Fix missing task alert emails on Dags using native template rendering #73228 - Fix missing task alert emails on Dags using native template rendering
• Stop providers lazy-loaded from writing the default config #73348 - Stop providers lazy-loaded from writing the default config
• Fix api-server --apps silently dropping apps and accepting typos #73371 - Fix api-server --apps silently dropping apps and accepting typos
• Fix Task SDK dry-run mode crashing on task instance start #73373 - Fix Task SDK dry-run mode crashing on task instance start
• Fix CLI --bundle-name finding no Dags in Git bundles #73377 - Fix CLI --bundle-name finding no Dags in Git bundles
• Fix empty audit log dag_id for DAG.cli() subcommands #73379 - Fix empty audit log dag_id for DAG.cli() subcommands
• Drop ignored --bundle-name and --dagfile-path from DAG.cli() #73387 - Drop ignored --bundle-name and --dagfile-path from DAG.cli()
• Fix indexing mapped task results with a NumPy integer #73400 - Fix indexing mapped task results with a NumPy integer
• Respect maximum_page_limit in batch list endpoints #73409 - Respect maximum_page_limit in batch list endpoints
• Stop the JWT revocation check stalling the API server event loop #73422 - Stop the JWT revocation check stalling the API server event loop
• Fix task instance 404 detail being a JSON array instead of a string #73424 - Fix task instance 404 detail being a JSON array instead of a string
• API: Remove unreachable Accept branch and last deprecated 422 alias #73463 - API: Remove unreachable Accept branch and last deprecated 422 alias
• API: Cap run_ids on the grid TI summaries stream endpoint #73465 - API: Cap run_ids on the grid TI summaries stream endpoint
• Rate-limit JWKS fetch attempts while the endpoint is down #73472 - Rate-limit JWKS fetch attempts while the endpoint is down
• UI: Fix grid columns vanishing when a Dag version has been cleaned #73514 - UI: Fix grid columns vanishing when a Dag version has been cleaned

These pull requests stay open because maintainers are already engaged in them - they count towards your limit:

• Fix scheduler refusing to start when [scheduler] only_idle is enabled #73141 - Fix scheduler refusing to start when [scheduler] only_idle is enabled
• Batch Team lookups when syncing Dag bundles to the database #73154 - Batch Team lookups when syncing Dag bundles to the database

This is not a judgement of you or of your changes. We never told contributors before that opening many pull requests at once was a problem, so there is nothing to feel bad about - and nothing is lost: your branches, commits and the review history stay where they are.

What we ask you to do is to make your first prioritization decision: choose which of the pull requests above matter most to you, and reopen them (up to 5 open at a time, including the ones still open) with the "Reopen pull request" button or gh pr reopen <PR_NUMBER> --repo apache/airflow. Reopen the ones you are ready to follow through - keep them rebased, respond to review comments and fix failing checks.

While your pull requests are waiting for review, the most valuable thing you can do is help in other ways - reviewing other contributors' pull requests, helping with issues, and taking part in the discussions on the devlist and Slack.

Why we introduced the limit, what it means for you and how to reopen or restore a pull request is explained in https://github.com/apache/airflow/blob/main/contributing-docs/32_open_pull_request_limit.rst.

---

Drafted-by: Claude Code (Opus 5); reviewed by @potiuk before posting