What this PR does / why we need it:

Ray offline store tests were running 3 times across different workflows, and Ray compute engine tests were running in the main CI instead of the dedicated Ray workflow.

• Removed the try/except block that conditionally appended RayDataSourceCreator to AVAILABLE_OFFLINE_STORES. This was the root cause - Ray was being auto-included in every workflow where it was installable.
• Changed online store from RedisOnlineStoreCreator to {"type": "sqlite"}. The compute engine tests focus on Ray compute, not online store behavior, so sqlite is sufficient and avoids needing redis in the pixi environment.
• Split the single test task into test-offline and test-compute, then combined them with depends-on. The Ray workflow now runs both sdk/python/tests/integration/offline_store (with Ray config) and sdk/python/tests/integration/compute_engines/ray_compute.
• Added --ignore=sdk/python/tests/integration/compute_engines/ray_compute to test-python-integration (main CI target).
• Replaced test-python-universal-ray-offline and test-python-ray-compute-engine with a single test-python-ray-integration target.