Fix consumption of CPU-backed interchange protocol dataframes

[shwina]

Description

Closes #11245. This PR fixes a bug in our code that consumes a protocol DataFrame that is backed by CPU memory.

This enables using the from_dataframe() function to construct DataFrames from other libraries:

In [12]: pdf = pd.DataFrame({'x': [1, 2, 3], 'y': [1.0, 2.0, np.nan], 'z': ['a', 'b', 'c']})

In [13]: vdf = vaex.from_dict({'x': [1, 2, 3], 'y': [1.0, 2.0, np.nan]})

In [14]: cudf.from_dataframe(pdf, allow_copy=True)
Out[14]: 
   x    y  z
0  1  1.0  a
1  2  2.0  b
2  3  NaN  c

In [15]: cudf.from_dataframe(vdf, allow_copy=True)
Out[15]: 
   x    y
0  1  1.0
1  2  2.0
2  3  NaN

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[github-actions]

This PR has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates. This PR will be labeled inactive-90d if there is no activity in the next 60 days.

[bdice]

@shwina This popped back up on my radar after you bumped the branch. If this is still active, I think it only needs a bit more work to be complete?

[wence-]

Some minor nitpicks, but otherwise looks good.

[wence-]

Is it deliberate that BYTEMASK is an int, but the other mask kinds are tuples? (They both turn into ints on construction, but the tuple instantiation looks kind of weird.

[wence-]

Style, I (at least) find double-negated conditions hard to parse. Perhaps have the truthy case in the if, and the exceptional case in the else branch:

if allow_copy:
    dbuf = ...
    return ...
else:
    raise TypeError(...)

[shwina]

Done

[ajschmidt8]

Removing ops-codeowners from the required reviews since it doesn't seem there are any file changes that we're responsible for. Feel free to add us back if necessary.

[vyasr]

@shwina what's the status of this PR?

[shwina]

Thanks @vyasr and apologies all for letting this PR go stale - I've updated the PR to target 23.06, and I believe addressed all the review comments.

@bdice, @wence-, would you be willing to take a final look? Thank you for the suggested improvements!

[wence-]

Implementation looks good, a minor request to add a little more testing in the pandas<->cudf interop tests

[wence-]

Suggested change

	cudf.from_dataframe(df, allow_copy=True)
	cdf = cudf.from_dataframe(df, allow_copy=True)
	assert_eq(df, cdf)

Add a case where there are nulls in the pandas frame?

Also can we check the round-trip in the opposite direction? (Assuming that is supported too).

[shwina]

Thanks! I added some null tests.

Also can we check the round-trip in the opposite direction? (Assuming that is supported too).

Today, that segfaults because Pandas doesn't check for the device type of the incoming buffers and tries to interpret the buffer pointers as host pointers :(

[bdice]

One minor change for docstring syntax, otherwise LGTM.

[wence-]

Thanks, apologies for the delay

[shwina]

/merge