Need to update calls to cuDF Series ctors when calling with a Buffer type

[rlratzel]

rapidsai/cudf#9370 changed support for howSeries object are to be constructed with Buffer objects.

For instance, this example is no longer supported:

import numpy as np
import cudf

tmp = cudf.Series([1, 2, 3])
print(tmp);

buffer = cudf.core.buffer.Buffer(tmp.data)
tmp2 = cudf.Series(data=buffer, dtype=np.dtype("int32"))
print(tmp2)

The supported pattern is now:

...
buffer = cudf.core.buffer.Buffer(tmp.data)
col = cudf.core.column.build_column(buffer, dtype=np.dtype("int32"))
tmp2 = cudf.Series._from_data({SERIES_NAME: col})

cc @ChuckHastings @vyasr

[vyasr]

@rlratzel @ChuckHastings I'm glad this was able to get resolved relatively quickly. As we discussed today I'll plan to be more conservative with future changes, but I think it would also be useful for us to meet at some point to get a sense of what parts of cuDF's API cuGraph might be using. There are many aspects that we would consider internal and may eventually want to hide, but some of them may need to stay public or at least stable for other RAPIDS libraries to use.

[ChuckHastings]

I'll yield to @rlratzel on most of these issues. He leads the python side of things.

I believe we are going to dramatically refactor/simplify much of our interface over the next few months. It seems like a natural time to review all points where we touch cudf as the python side is refactored. I think much of what is there was copied from python/cython in cudf at the time when we needed to implement something. I have no doubt that there are more examples where we are doing things using techniques that are not preferred.

[vyasr]

OK sounds good @rlratzel I don't know if we necessarily need to have a dedicated conversation for this, that's up to you (perhaps after you've planned more of your refactoring). I'll keep alerting you (and marking relevant PRs) as we make further internal changes in cuDF that may impact you.