Fix cuvs build due to rmm and cuCollections api changes

[jcrist]

rapids-cmake just bumped the commit used for cuCollections in builds, leading to cuvs build failures.

Here we fix usage of cuco::default_filter_policy that were changed in NVIDIA/cuCollections#827.

Also fixes some fallout from rapidsai/rmm#2527, merging in #2503 (as well as a few small fixups).

Closes #2503.

[jcrist]

For some reason local builds are failing for me in some raft compilation errors that we haven't seen in CI. This is on a clean fork and a conda environment built from conda/environments/all_cuda-129_arch-x86_64.yaml, so I'm not exactly sure what's going wrong. Hopefully things pass fine in CI.

[jcrist]

Hmmm, ok, also seeing the same failures here. At least I know it's not just my env then!

Looks like some other upstream change has also broken things, beyond just the cuco bump.

[jcrist]

Ah, it's rapidsai/rmm#2527, which broke some code in raft.

[wence-]

Ah, it's rapidsai/rmm#2527, which broke some code in raft.

Just FWIW, that code in raft was always wrong, sorry :(

[jcrist]

Build failures due to raft should be fixed by NVIDIA/raft#3123

[jcrist]

I've folded in the changes from #2503 as well, since both PRs will be needed to fix the cuvs builds. I also had to push a small fixup there to handle a few missed cases.

[jameslamb]

/merge