feat: add public dataset downloader and cache guard

[edithatogo]

Summary

Add reusable local-only downloader/cache tooling for public datasets used by worked examples, with guardrails that prevent raw patient-level files from being committed.

Linked Conductor track

conductor/tracks/public_clinical_dataset_worked_example_20260704

Why this is separate

The initial worked example can provide local-path instructions and fixtures. A reusable downloader/cache guard affects multiple datasets and should have its own security and data-governance validation.

Acceptance criteria

• Support manifest-driven local download/cache verification where the dataset license permits local use.
• Never commit raw public clinical dataset files.
• Add commit/CI guard fixtures for accidental raw dataset paths and known MIMIC table names.
• Emit clear diagnostics for missing files, license acknowledgement, and cache integrity.

Dependencies

• Public Clinical Dataset Worked Example manifest and local-cache policy.
• Existing licensed asset guard and source/license audit patterns.