Respect async connection hook overrides

[bobu-putheeckal]

Fixes: #66845

This PR fixes a narrow async hook dispatch issue in provider connection handling.

Problem

airflow.providers.common.compat.connection.get_async_connection() always resolved connections through BaseHook. That means async hooks which intentionally override aget_connection() or get_connection() on a subclass can be bypassed when code reaches the shared async helper.

The issue report calls this out for async provider hooks: subclass-specific connection resolution should still be honored in the async path.

What changed

• get_async_connection() now accepts an optional hook_class argument.
• The default remains BaseHook, preserving behavior for existing callers.
• HttpAsyncHook.config() passes type(self) so subclass aget_connection() / get_connection() overrides are reached through normal classmethod dispatch.
• Added focused regression coverage in the common compat provider and HTTP provider tests.

Validation

I ran these checks locally:

uv run --project providers/common/compat --group dev pytest providers/common/compat/tests/unit/common/compat/connection/test_connection.py
# 4 passed

uv run --project providers/http --group dev pytest providers/http/tests/unit/http/hooks/test_http.py::TestHttpAsyncHook::test_config_uses_subclass_aget_connection
# 1 passed

uv run --project providers/http --group dev ruff format providers/common/compat/src/airflow/providers/common/compat/connection/__init__.py providers/common/compat/tests/unit/common/compat/connection/test_connection.py providers/http/src/airflow/providers/http/hooks/http.py providers/http/tests/unit/http/hooks/test_http.py
# 4 files unchanged

uv run --project providers/http --group dev ruff check providers/common/compat/src/airflow/providers/common/compat/connection/__init__.py providers/common/compat/tests/unit/common/compat/connection/test_connection.py providers/http/src/airflow/providers/http/hooks/http.py providers/http/tests/unit/http/hooks/test_http.py
# all checks passed

No UI behavior is changed, so screenshots are not applicable.

---

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

• Yes: OpenAI Codex. I reviewed the generated changes and ran the focused tests listed above.

Generated-by: OpenAI Codex following https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions

[kaxil]

AI Spam! Review your PR descriptions!

[bobu-putheeckal]

Thanks, you are right that the previous description was not good enough. I rewrote it to describe the concrete bug, the code path changed, and the exact local validation results. I will slow down and avoid batch-style Airflow submissions going forward.

[bobu-putheeckal]

I also moved this to draft so it is not sitting in the review queue while I make sure the description and evidence meet the project expectations.

[potiuk]

@bobu-putheeckal — closing this for now since it's a draft that's been inactive for a few weeks. This is purely housekeeping to keep the review queue tidy — not a reflection on the change itself.

When you're ready to pick it back up, just reopen this PR (or open a fresh one), rebase onto the latest main, and mark it Ready for review. There's no rush, and thanks for the contribution!

See the PR quality criteria.

_{Automated triage note drafted by an AI-assisted tool — may get things wrong; a real Apache Airflow maintainer is happy to help if you reopen. (why automated)}

---

Drafted-by: Claude Code (Opus 4.8); reviewed by @potiuk before posting