Parquet output requires `pip install pyarrow` (not auto-installed)

[sqllocks]

Issue

On a fresh clean-machine install (pip install sqllocks-spindle v2.13.0 + Python 3.12), running:

spindle generate retail --scale small --seed 42 --output ./demo-data/ --format parquet

…fails with:

ImportError: Unable to find a usable engine; tried using: 'pyarrow', 'fastparquet'.
A suitable version of pyarrow or fastparquet is required for parquet support.

Workaround: pip install pyarrow — then it works perfectly.

Why this matters

Spindle's pitch is Microsoft Fabric / Lakehouse, which writes Delta/Parquet. The README quick-start mentions Lakehouse, Warehouse, etc. — but the default install can't write parquet without an extra step.

Recommended fixes (pick one)

1. Add pyarrow to core install_requires — simplest, but adds ~50MB to the install
2. Add a [fabric] or [parquet] extras_require — users can opt-in via pip install sqllocks-spindle[fabric]
3. Document clearly in README — explicit "for parquet output, also install pyarrow"

Leaning toward #2 — keeps the core install small but makes the Fabric story one command:

pip install sqllocks-spindle[fabric]

Test repro

Fresh venv, Python 3.12.10, Windows. Full session captured in tracker.

Friction context

Found during the Step 10 launch-pre-flight test (clean-machine install). End-to-end time from venv creation to working CSV output was 22 seconds. Parquet was the only blocker.