feat(sample-data): list-driven sample data generation for --test

[Xander-git]

Summary

Replaces the single hardcoded --add-test-samples equipment literal with a list-driven sample-data generator. A declarative SAMPLES list (2 equipment / 3 projects / 7 runs) is expanded into a full on-disk demo tree under the -test sandbox, with every folder's metadata written through the real production producers (ReadmeGenerator, CreationWriter, EquipmentCacheWriter) so it can't drift from what the wizard writes.

Design spec: docs/superpowers/specs/2026-05-29-sample-data-generation-design.md.

What changed

• sample_data/ — spec.py (Pydantic SAMPLES), generator.py (SampleDataGenerator + generate_samples() facade + guardrailed destructive wipe), __init__.py.
• controller/metadata_assembly.py — extracted build_readme_context / build_creation_json (+ moved pure helpers) now shared by the controller and the seeder. controller/creation.py delegates to them (behavior-preserving, guarded by parity tests).
• config/test_bootstrap.py — shared write_starter_test_config / bootstrap_test_config in a light module (no pystray import), used by both the tray flag and the dev command.
• dev/seed.py — python -m exlab_wizard.dev.seed (always wipe + regenerate).
• tray/main.py — deletes the inline TESTRIG literal; seeds via the shared bootstrap on first launch only (never wipes an existing sandbox).

Design notes

• Determinism: fixed clock; README/creation.json/payload bytes are byte-identical across runs. Only equipment.json's writer-stamped first_seen_at/last_modified_at differ (documented; excluded from the determinism assertions).
• Faithfulness: seeded EquipmentConfig.local_root/nas_root are base roots so run-walking (orchestrator quiescence poller / validator, which compose local_root / equipment.id) discovers the tree.
• Safety: the destructive wipe is fenced to the -test sandbox (_app_name() ends in -test) with strict-subpath containment; only ever removes local_root/<prefixed_id>. Symlink-escape and out-of-sandbox cases are tested.
• Seeded SYNCED / BLOCKED_BY_VALIDATION values are deliberate UI fixtures written into creation.json; no run is actually verified or transferred.

Test plan

• uv run pytest tests/unit tests/integration --ignore=tests/integration/test_pyinstaller_smoke.py → 2362 passed, 3 skipped (44 new tests: assembly parity, generator e2e, determinism, guardrails incl. symlink escape, base-root fidelity, dev command, tray no-op).
• ruff check / ruff format --check / mypy clean on all changed files.
• Real python -m exlab_wizard.dev.seed run produces the expected tree (2 equipment, 3 projects, 7 runs, 10 metadata triples, 2 equipment.json, 3 test_runs.json, 1 scan-trigger payload) and is byte-deterministic across two runs.

🤖 Generated with Claude Code