feat: add dry-run flag to build_dataset

[Scr4tch587]

What

Add a dry_run flag to build_dataset that selects the store and returns the produced rows.

How

• build_dataset(dry_run=...) is the single boundary that reads the flag
• real build → PostgresStore, returns None
• dry run → fresh MemoryStore, returns the produced rows for the requested dataset
• MemoryStore starts empty, so a dry run rebuilds the whole graph in isolation
• no build lock and no cleanup — store is request-private and GC'd with the request

Why

Validate builder output end-to-end without writing to or reading from the real DB.

[github-actions]

• main
  • feat: add Store abstraction for build data access #352
    • feat: add dry-run flag to build_dataset 👈
      • feat: add dry-run query param to build endpoint #354
        • docs: document dry-run builds in backend spec #355