Support raw datastore-path folders as command inputs

[corcra]

🤖 This PR was authored by an AI agent (GitHub Copilot CLI, model Claude Opus 4.8) working with @corcra. Please review accordingly.

Closes #14

Summary

Inputs previously accepted only registered data assets (alias=name[:version]) and prior-job output dirs (alias=job_dir:<job_id>:<path>). Outputs already supported raw datastore folders (alias=datastore/folder). This adds the symmetric input branch so jobs can mount/download an existing datastore folder directly, without pre-registering a data asset.

Changes

• data.py: add _is_alias_datastore_path_string predicate and a datastore-path branch in _get_data_assets that builds an azureml://datastores/<ds>/paths/<folder> Input. Branch order is job_dir: → datastore-path → data-asset.
• data.py: extract shared _datastore_uri helper, now used by both inputs and outputs.
• __main__.py: document the alias=datastore/folder form in --download/--mount help.
• docs/examples.md: add CLI + Python examples for the new input form.
• tests/test_data.py: parametrized predicate tests; input-routing tests asserting the resulting URI/mode and that ml_client.data.get is not called for datastore paths but is for name:version.

Design notes

• Disambiguation: a / in the RHS reliably signals a datastore path (AML data-asset names can't contain /); name:version and job_dir: are matched without relying on /.
• The predicate is a pure string check on purpose — _extract_alias_datastore_path calls sys.exit(1) rather than raising, so it can't be wrapped in try/except.
• Tradeoff (opt-in): a raw datastore path is mutable and unversioned, so runs record no data-asset version/lineage — identical to how the output datastore form already behaves. Users wanting versioning keep using name:version (unchanged).

Verification

ruff clean, ty clean, 80 passed.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[github-actions]

📖 Docs preview: https://smokeshow.helpmanual.io/6j2f3o0w0m384m3r2224/

[Copilot]

Pull request overview

Adds symmetric support for using raw datastore folder paths as command inputs (in addition to existing output support), allowing users to mount/download alias=datastore/folder/... without registering a data asset.

Changes:

• Add _is_alias_datastore_path_string plus a new datastore-path routing branch in _get_data_assets to construct azureml://datastores/<ds>/paths/<folder> inputs.
• Factor shared datastore URI formatting into _datastore_uri, used by both inputs and outputs.
• Update CLI help and documentation, and add tests to validate routing and ensure ml_client.data.get is bypassed for datastore paths.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/submit_aml/data.py	Adds datastore-path detection and input construction; extracts _datastore_uri and reuses it for outputs.
tests/test_data.py	Adds predicate tests and input-routing tests (including “no AML lookup for datastore paths”).
src/submit_aml/__main__.py	Updates --download/--mount help text to document alias=datastore/folder input form.
docs/examples.md	Adds CLI + Python examples demonstrating mounting a raw datastore folder as an input.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[fepegar]

Closing in favour of #17 as discussed offline.