ml/azure-ai-ml: audit and improve e2etest skip/live_test_only marker documentation

[Copilot]

All @pytest.mark.skip and @pytest.mark.live_test_only markers across ~35 e2etest directories under sdk/ml/azure-ai-ml/tests/**/e2etests were audited. Bare markers and vague reasons have been replaced with actionable documentation.

Changes

Bare @pytest.mark.skip() → documented reasons

All @pytest.mark.skip() calls without a reason= now include one, covering Docker daemon requirements (local endpoint tests), infrastructure unavailability, and API removal.

Vague reasons replaced with root causes

• "investigate later", "Skip for not ready." → specific explanation of the blocking condition
• "Flaky test" → root cause (HTTP timeout under load, no retry logic)
• "request header size being too large" → prerequisite context (materialization store must be enabled on sdk_vnext_cli_fs)

ADO work item references standardized

Internal Azure DevOps IDs prefixed with ADO for clarity:

# Before
@pytest.mark.skip(reason="Task 1791832: Inefficient...")
# After
@pytest.mark.skip(reason="TODO (ADO 1791832): The full-model-list operation is slow enough to cause pipeline timeouts...")

Recurring skip patterns — improved across all instances

Pattern	Files affected	Root cause added
"parallel e2e recording issue"	8+ markers in dsl/pipeline	Non-deterministic parallel component step ordering in test-proxy recordings
"Spark version Upgrade"	6+ markers	Spark version incompatibility in test environment
TODO (2370129)	dsl + pipeline	"Cannot find pipeline run" recording ID mismatch
TODO (2177353)	dsl + pipeline + control_flow	Non-deterministic component version ordering between recording and playback

Specific test skips clarified

• test_promote_model: _prepare_to_copy was removed from the SDK; test must be rewritten to use MLClient.models.share()
• Monitoring tests (test_monitor_schedule.py): iris-endpoint resources were deleted from the test workspace; tests need a replacement endpoint
• test_model_archive_restore_container: uses client.models.list() on every assertion, causing pipeline timeouts (ADO 1791832)

Not addressed (out of scope)

11 @pytest.mark.live_test_only tests across 5 files remain unchanged. These have stale recordings that no longer match current sanitizer patterns. Re-enabling them requires:

1. A live Azure ML workspace with AZURE_TEST_RUN_LIVE=true
2. Re-running the tests to generate new recordings
3. Pushing recordings to Azure/azure-sdk-assets (tag: python/ml/azure-ai-ml_d3ebe00cb3)
4. Removing the live_test_only marker

Affected files: test_connections.py, test_datastore.py, test_environment.py, test_component.py, test_dsl_pipeline_samples.py.

[azure-pipelines]

Azure Pipelines:
10 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.