fix(super_component): only expose run_async when wrapped pipeline is AsyncPipeline

[alvinttang]

Summary

SuperComponent declared async def run_async(...) at the class level, so ComponentMeta.__call__'s post-__init__ hasattr(instance, \"run_async\") check set __haystack_supports_async__ = True for every instance — including those wrapping a sync Pipeline. Two visible failures resulted:

1. Calling wrapper.run_async(...) on a sync-pipeline-backed SuperComponent raised TypeError: Pipeline is not an AsyncPipeline....
2. Putting such a SuperComponent inside an outer AsyncPipeline made AsyncPipeline._run_component_async dispatch to run_async, and the whole outer pipeline crashed with PipelineRuntimeError.

Fixes #9435

Fix

haystack/core/super_component/super_component.py:

• Renamed the class-level method from run_async to _run_async. The metaclass no longer sees a class-level run_async and therefore no longer marks every SuperComponent as async-capable.
• __init__ binds self.run_async = self._run_async per-instance only when isinstance(pipeline, AsyncPipeline). Sync-pipeline-backed instances simply do not have a run_async attribute.
• Added a TYPE_CHECKING stub for run_async so the static typing surface (used by the existing tests and downstream users) is preserved while the runtime attribute is bound lazily.

Test

test/core/super_component/test_super_component.py:

• test_sync_pipeline_does_not_advertise_async_support — __haystack_supports_async__ is False and run_async attribute is absent.
• test_async_pipeline_advertises_async_support — positive control.
• test_super_component_with_sync_pipeline_inside_async_pipeline — the regression scenario, runs outer.run_async(...) and verifies it succeeds.

All three RED before the patch, GREEN after. Existing tests test_super_component_run_async and test_super_component_async_serialization_deserialization still pass.

Locally on darwin/arm64:

• pytest test/core/super_component/ test/core/component/ → 123 passed.
• Three pre-existing failures in test_pipeline_base.py (test_show_in_notebook, test_find_super_components, test_merge_super_component_pipelines) are unrelated and reproduce on main without this patch.

Risk notes

hasattr(super_component, \"run_async\") now returns False for sync-pipeline-backed SuperComponents. Any caller that previously hit the broken run_async (which always raised TypeError) will now hit AttributeError instead — strictly an improvement. The @super_component decorator path goes through _SuperComponent.__init__ so it inherits the fix automatically.

[vercel]

Someone is attempting to deploy a commit to the deepset Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

alvinttang seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[julian-risch]

Hi @alvinttang Thank you for opening this PR. We need you to agree to the CLA please. In particular, this commit was not done by a GitHub user: 73be582 Could you fix that please?