[BUG] librmm conda packages are shipping dependencies in the package

[bdice]

Currently, librmm is shipping headers and libraries from its dependencies. We need to clean this up so we don't clobber.

This issue is similar in flavor to rapidsai/cudf#13230.

I think this should be a prerequisite for #1779, to prevent the problem from getting worse.

rapids-logger

#1528 fixed librmm's clobbering of spdlog and fmt headers. However, we are now shipping rapids-logger headers and .so files that should just come from including rapids-logger as a dependency. This is a simple fix for the recipe, which I filed in #1834.

NVTX

For conda builds, NVTX should come from nvtx-c conda packages, rather than being CPM-fetched and shipped in include/nvtx3/ in the librmm package (which clobbers nvtx-c!). However, there is no CMake config being shipped with NVTX3 upstream, so we probably need to fix that first -- or provide our own CMake Find module so that it can be found locally. This may be a rapids-cmake project or a project for upstream NVTX. I filed an issue here: NVIDIA/NVTX#113

edit: This is fixed by #1930.

CCCL

We are shipping a vendored CCCL in librmm. We should probably transition this to use rapids_core_dependencies as proposed in #1307. I verified that this should contain the right files for our RAPIDS-vendored-CCCL installation, so the fix should be straightforward. The latest plan is here: NVIDIA/cccl#1939 (comment)

[vyasr]

Re:CCCL, see #1307 (comment).

[bdice]

Thanks @vyasr -- I have a revised plan for CCCL here: NVIDIA/cccl#1939 (comment)