[BUG] handling of bytes cols

[lmeyerov]

Currently, a bytestring col in api=3 returns the following non-obvious error:

Exception: {'data': {'args': {'compression': None, 'dataset_id': 'ef83f4f0ecb442e082d0edfb974f5f95'}, 'error_message': 'cuDF failure at: /conda/conda-bld/libcudf_1591199195844/work/cpp/src/column/column_view.cpp:48: Compound (parent) columns cannot have data'}, 'error_code': 400, 'message': 'Failed to parse', 'success': False}

At a minimum, we should give a more friendly error. Even better, we may be able to detect & coerce to str in the backend. Finally, should probably check w/ cuDF on expected behavior.

Originally brought up by Vinayaka

[lmeyerov]

This may really be two things:

1. PyGraphistry client: Warn on unexpected dtypes, and potentially add a default-off coercion mode

2. Same-but-different for the upload endpoints: by default, clear error message on unexpected dtypes, and default-off option to auto-coerce

Officially, we should aim to support the full set of RAPIDS-supported dtypes: https://docs.rapids.ai/api/cudf/nightly/basics.html#supported-dtypes