feat: add SetsClient for persistent saved sets

[gtchax]

Summary

Adds a SetsClient to the Python SDK so users (and DataMaker sandbox scripts) can manage sets — persistent, named snapshots of generated/fetched rows (e.g. a pinned regression dataset or a shared fixture). This is the SDK half of the new Sets feature in the main DataMaker app; the /sets CRUD API it talks to already exists server-side.

What's included

• routes/sets.py → SetsClient with the full CRUD surface:
  • get_sets(project_id=None) — list (scoped by project; falls back to DATAMAKER_PROJECT_ID)
  • get_set(set_id) — fetch one, including its full data payload
  • create_set(name, data=None, description=None, row_count=None, project_id=None)
  • update_set(set_id, ...) — sends only provided fields
  • delete_set(set_id)
  • save_set(name, data, description=None, project_id=None) — convenience for the common "save these rows as a named set" flow, pulling project context from DATAMAKER_PROJECT_ID when omitted
• Wired into routes/__init__.py and main.py (delegating methods on the top-level DataMaker client + a .sets property for direct access), mirroring the existing resource clients.
• TestSetsClient unit tests covering list (scoped/unscoped), get, create (full/minimal/env-fallback/validation), update, delete, and save_set.
• A Set Management section in llms.txt.

Usage

dm = DataMaker()
rows = dm.generate(template)
result = dm.save_set("nightly-regression", rows)
print(f"Saved set: {result['name']} ({result['rowCount']} rows)")

Testing

Full unit suite green locally — 62 passed, 24 skipped (integration tests skip without a live API key). ruff format clean on the new/changed files.

🤖 Generated with Claude Code