What this PR does / why we need it:

• tests requiring container resources in test_universal_registry.py are pinned to a single worker using xdist_group to cut down on setup times as session-scoped fixtures aren't reused across workers.
• test_universal_cli.py was unnecessarily marked with universal_offline_stores, changed to run only once with file offline store.
• test_apply_feature_view_integration and test_apply_entity_integration tests in test_feature_store.py are removed as there are basically identical to test_apply_feature_view_success and test_apply_entity_success tests in test_universal_registry.py.
• test_remote.py was removed as it was testing for only a small subset of functionality compared to test_universal_registry.py. mock_remote_registry fixture from test_remote.py was instead moved to test_universal_registry.py. This was impossible before because universal registry tests depended on apply/delete operations as well.
• test_registry_server.py is also effectively superseded by test_universal_registry.py as it's testing for a much wider range of functionality from a higher-level interface (remote registry instead of direct grpc server calls).

P.S. In my dev environment the changes cut the overall runtime of test-python-integration-local from ~7m down to ~4m.