Test failures present in MG bulk sampler tests

[rlratzel]

The test run was on a 2-node 16-GPU (total) configuration:

=========================== short test summary info ============================
FAILED tests/sampling/test_bulk_sampler_mg.py::test_bulk_sampler_simple - Run...
FAILED tests/sampling/test_bulk_sampler_mg.py::test_bulk_sampler_mg_graph_sg_input
================== 2 failed, 1 skipped, 3 warnings in 51.13s ===================

Both tests had the same problem:

        bs.flush()

>       recovered_samples = cudf.read_parquet(os.path.join(tempdir_object.name, "rank=0"))

tests/sampling/test_bulk_sampler_mg.py:63:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
/usr/local/lib/python3.8/dist-packages/nvtx/nvtx.py:101: in inner
    result = func(*args, **kwargs)
/usr/local/lib/python3.8/dist-packages/cudf/io/parquet.py:550: in read_parquet
    return _parquet_to_frame(
/usr/local/lib/python3.8/dist-packages/nvtx/nvtx.py:101: in inner
    result = func(*args, **kwargs)
/usr/local/lib/python3.8/dist-packages/cudf/io/parquet.py:611: in _parquet_to_frame
    _read_parquet(
/usr/local/lib/python3.8/dist-packages/nvtx/nvtx.py:101: in inner
    result = func(*args, **kwargs)
/usr/local/lib/python3.8/dist-packages/cudf/io/parquet.py:688: in _read_parquet
    return libparquet.read_parquet(
parquet.pyx:123: in cudf._lib.parquet.read_parquet
    ???
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

>   ???
E   RuntimeError: CUDF failure at: /project/cpp/src/io/parquet/reader_impl_helpers.cpp:262: All sources must have the same number of columns

parquet.pyx:182: RuntimeError