[FEA] Use RMM with FAISS GPUResources

[cjnolet]

Recent updates to FAISS now provide a means to plug in our own custom memory manager (RMM) so that we can control each allocation and deallocation.

This is needed on cuml, ideally before the approximate indices are included, since they require indexes with memory exclusively owned and managed by FAISS. This is particularly a concern when using the pool allocator, because a user could allocate most of their GPU's memory for RAPIDS RMM and be confused when they run KNN with approximate indices and receive out of memory errors.

[cjnolet]

This is currently blocked by conda-forge/faiss-split-feedstock#17. We need to wait for the FAISS 1.6.4 conda package

[h-vetinari]

Just FYI, conda-forge/faiss-split-feedstock#17 has been merged 🙃

[cjnolet]

This change will also require the hardcoded int64_t knn indices to be updated to int everywhere. We should be able to use the int64_t indices when necessary (such as in scaling the mnmg knn).

[JohnZed]

Particularly important for approximate NN, in which FAISS owns a larger block of memory. Will also require updating indices back to 32-bit for newer FAISS rev.

[github-actions]

This issue has been marked stale due to no recent activity in the past 30d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be marked rotten if there is no activity in the next 60d.

[github-actions]

This issue has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed.