Document how to use a GCP service account in Airflow with SkyPilot

[kevinmingtarja]

This PRs adds an example for setting up a SkyPilot task in an Airflow pipeline to use a GCP service account stored as an Airflow connection.

Main changes:

• Switch to @task.virtualenv to run the sky SDK calls in its own virtual environment. This was mainly to get around a dependency conflict, as Airflow is not compatible with sqlalchemy 2.0 (Fix all deprecations for SQLAlchemy 2.0 apache/airflow#28723)
• Related, removed task_failure_callback, because sky commands can only work inside the venv, and on_failure_callback functions cannot be run inside a venv. But anyways I think this is fine, because we set down=True when calling launch, so the SkyPilot clusters will automatically spin down once the task finishes (regardless of success/failure)
• Fetch GCP service account JSON key from an Airflow connection, and mount it on the SkyPilot cluster
• Added data_preprocessing_gcp_sa.yaml (also replicated to add example yaml for using dynamically mounted GCP service accounts mock-train-workflow#4)
• Added a new screenshot, and updated all screenshots based on the latest Airflow UI, for consistency

Manually tested using a Remote SkyPilot API server running on GKE and Airflow running locally.

Tested (run the relevant ones):

• Code formatting: install pre-commit (auto-check on commit) or bash format.sh
• Any manual or new tests for this PR (please specify below)
• All smoke tests: /smoke-test (CI) or pytest tests/test_smoke.py (local)
• Relevant individual tests: /smoke-test -k test_name (CI) or pytest tests/test_smoke.py::test_name (local)
• Backward compatibility: /quicktest-core (CI) or pytest tests/smoke_tests/test_backward_compat.py (local)

[romilbhardwaj]

Thanks @kevinmingtarja!