Revert version-drift change (#34) to restore working baseline

[klagrida]

Why

PR #34 bumped Spark (3.5.0→4.0.1), Airflow (3.0.2→3.1.7), the Iceberg Spark runtime, and the Airflow image. Since landing on main it has kept CI red: the larger images saturate the single-node minikube's Docker Hub pull bandwidth, so the Airflow Helm release and the Jupyter/monitoring rollouts time out non-deterministically (same deploy ran 6m17s in one run and >25min in another). Timeout-bumping (PRs #37/#38) proved to be whack-a-mole rather than a fix.

This restores the last known-good state — commit 1fe2033 (Dec 20), the parent of #34.

What

git revert 540ea5e. The resulting tree is byte-identical to 1fe2033 for the 7 affected files (git diff 1fe2033 HEAD is empty):

• Spark 4.0.1 → 3.5.0
• Airflow chart 1.19.0 → 1.18.0, appVersion/tag 3.1.7 → 3.0.2
• Dockerfile apache/airflow:3.1.7 → 3.1.5-python3.13
• Iceberg runtime 4.0_2.13 → 3.5_2.12, hadoop-aws 3.4.1 → 3.3.4
• Restores the Hive config lines that Fix version drift across Terraform/Docker and remove leftover Hive config #34 also removed

Follow-ups

The version-upgrade + Hive cleanup intent is still tracked. Re-attempting it should be paired with the durable CI image-pull fix (issue #36: pre-pull/cache images into minikube, pin a smaller Jupyter image) so the heavier images don't re-break CI.

🤖 Generated with Claude Code