Add libcudf wrappers around current_device_resource functions.

[harrism]

Description

Merge after rapidsai/rmm#1661

Creates and uses CUDF internal wrappers around RMM current_device_resource functions.

I've marked this PR as breaking because it breaks the ABI, however the API is compatible.

For reviewers, the most substantial additions are in the new file <cudf/utilities/memory_resource.hpp>, and in the DEVELOPER_GUIDE.md and *.rst docs. The rest are all replacements of an include and all calls to rmm::get_current_device_resource() with cudf::get_current_device_resource_ref().

Closes #16676

Checklist

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

[harrism]

@vyasr I could use your help understanding the docs build failure. Thanks!

[vyasr]

You'll need to add a new docs page for memory_resource.hpp to https://github.com/rapidsai/cudf/tree/branch-24.10/docs/cudf/source/libcudf_docs/api_docs, and then add it either to index.rst or to one of the subpages that it lists in its ToC (or their subpages, etc).

[harrism]

You'll need to add a new docs page for memory_resource.hpp to https://github.com/rapidsai/cudf/tree/branch-24.10/docs/cudf/source/libcudf_docs/api_docs, and then add it either to index.rst or to one of the subpages that it lists in its ToC (or their subpages, etc).

Thanks! Made an attempt.

[nvdbaranec]

This may have been asked already, but why do we have to include <cudf/utilities/default_stream.hpp> as well as <cudf/utilities/memory_resource.hpp> ? Or were the few places where this was done because it wasn't explicitly included for the neighboring cudf::get_default_stream?

[nvdbaranec]

There are a handful of places where the default stream header is included by ultimately unused. Example:

cudf/cpp/include/cudf/reduction/detail/segmented_reduction_functions.hpp

Line 22 in f83b64d

#include <cudf/utilities/default_stream.hpp>

Edit: Actually I think this may be the only one.

[harrism]

This may have been asked already, but why do we have to include <cudf/utilities/default_stream.hpp> as well as <cudf/utilities/memory_resource.hpp> ? Or were the few places where this was done because it wasn't explicitly included for the neighboring cudf::get_default_stream?

Yes, I fixed IWYU mistakes when I noticed them.

[nvdbaranec]

Just that one lingering unnecessary header - not a blocker.

[davidwendt]

Looks like this include can be removed.

[davidwendt]

Looks like this include can be removed.

[harrism]

Correct, as well as device_memory_resource.hpp, but need to add rmm/resource_ref.hpp.

[harrism]

/merge