Multi-GPU out of core KMeans progress tracking issue

[viclafargue]

This issue aims at tracking progress on the multi-GPU out of core KMeans implementation and benchmarking.

Please tell me if anything was forgotten. Also, if you add a new issue/PR, please list it here and link this issue as its parent so we can keep track of everything that is going on.

Main Tasks

• Multi-GPU batched KMeans core (Multi-GPU Batched KMeans #2017)
• Multi-GPU KMeans implementations merge + cleanup ([Cleanup] Combine Batched and Regular KMeans Impl #2015)
• SNMG Batched KMeans Python API (SNMG Batched KMeans Python API #2154)
• Handle multiple partitions per rank/worker + API merges ([FEA] Multi-node Out of Core Streaming KMeans API #2066)
• Handle multiple partitions per rank/worker (cuML side) ([FEA] Scaling MNMG KMeans cuml#8201 and [FEA] Skip Device Concat for KmeansMG cuml#8084)
• Evaluate KMeansPlusPlus Init For Out of Core (Evaluate KMeansPlusPlus Init For Out of Core #2224)
• SNMG Batched KMeans Python API benchmark (SNMG Batched KMeans Python API and benchmarking #2149)
• Large scale multi-node Dask KMeans benchmark (Large scale multi-node Dask KMeans benchmark cuml#8198)

Improvements and bugfixes

• Out-of-core K-means improvements ([FEA] Out-of-core K-means improvements #2292)
• Apply ABI Breaking Changes for KMeans (Apply ABI Breaking Changes for KMeans #2147)
• Reuse Precomputed Norms for Inertia Computation ([FEA] cluster_cost can accept X norm #2057)
• Evaluate Prefetch for Batched Kmeans ([FEA] Evaluate Prefetch for Batched Kmeans #1917)
• Clamp max_iter in OOC KMeans test ([CI] Update Convergence Criteria to Fix Flaky KMeans Test #2256)
• KMeans ignores sample_weight when computing inertia and score ([BUG] KMeans ignores sample_weight when computing inertia and score cuml#8530)
• KMeans.transform returns squared distances instead of Euclidean distances ([BUG] KMeans.transform returns squared distances instead of Euclidean distances cuml#8536)