Fix silent no-op download of named job outputs in azure-ai-ml

[Copilot]

MLClient.jobs.download(name=..., output_name=...) downloaded nothing — and raised nothing — for a job's uri_folder/uri_file/mltable outputs. RunHistory reports a job output's type in PascalCase (UriFolder), but after the 1.35.0 msrest → arm_ml_service migration get_job_output_uris_from_dataplane() filtered those types against the new enum whose values are snake_case (uri_folder), so every data output was dropped before the dataplane was ever queried.

>>> client.jobs._runs_operations.get_run_data(job_name).run_metadata.outputs["forecast_data"].type
'UriFolder'
>>> from azure.ai.ml._restclient.arm_ml_service.models import DataType
>>> 'UriFolder' in [o.value for o in DataType]   # pre-fix filter
False

Changes

• _job_ops_helper.py — added _normalize_asset_type() (lower() + strip _) and precomputed _DATA_ASSET_TYPES / _MODEL_ASSET_TYPES sets; the dataset and model filters now match either wire spelling. Unknown and missing types are still skipped, so no dataplane calls are made for them.
• _job_ops_helper.py — read the model dataplane response via values_property; model_uris.values resolved to the inherited MutableMapping.values method and raised AttributeError: 'function' object has no attribute 'items'.
• _model_dataplane_operations.py — build the request as BatchGetResolvedUrisDto(values_property=model_ids); values= raised TypeError: unexpected keyword argument 'values', so model outputs failed even once they passed the filter. Serialized body is unchanged ({"values": [...]}) and now matches DatasetDataplaneOperations.
• _job_operations.py — when an explicitly requested output_name cannot be resolved, log at warning level instead of debug. Other misses (logs, default artifact store, batch scoring file) stay at debug.
• Tests — test_job_ops_helper.py covers both spellings for data and model outputs, a single named output, unknown/missing types, and the asset ids sent to both dataplane clients (via the real operations classes over mocked service clients); test_job_operations.py covers the new warning.
• CHANGELOG — entries under 1.36.0.

Notes for reviewers

• The two model-output bugs are separate from the reported uri_folder symptom but sit in the same resolution path and share the same migration root cause; without them, model outputs would still fail after the filter fix.
• stream_logs_until_completion's default_output.job_output_type == DataType.URI_FOLDER check was left alone: both sides come from the control plane (JobOutputType.URI_FOLDER == DataType.URI_FOLDER == "uri_folder").
• azure-ai-ml's dependencies could not be installed in the agent sandbox, so the repo test suite was not run locally; the code paths were exercised against the real generated models through a standalone harness, and CI covers the rest.

• Fixes Download of URI folder is broken in 1.35.0 #48941

[azure-pipelines]

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