[BUG] C++ benchmarks memory leak

[tfeher]

Describe the bug
Memory utilization continuously grows while running some of the C++ benchmarks, leading to OOM

Steps/Code to reproduce bug

make .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX -GNinja
ninja -j 40 install
bench/prims_benchmark

Output

...
DistanceCosineF/10/manual_time                                            25.6 ms         25.7 ms           27
DistanceCosineF/11/manual_time                                            50.0 ms         50.1 ms           14
DistanceCosineF/12/manual_time                                            3.63 ms         3.65 ms          193
terminate called after throwing an instance of 'rmm::out_of_memory'
  what():  std::bad_alloc: out_of_memory: CUDA error at: /opt/conda/envs/rapids/include/rmm/mr/device/cuda_memory_resource.hpp:70: cudaErrorMemoryAllocation out of memory
Aborted (core dumped)

Might be related to particular benchmarks. The following lead to OOM:

• bench/prims_benchmark --benchmark_filter=Gram*
• bench/prims_benchmark --benchmark_filter=Distance*

Expected behavior
Memory consumption within bounds.

Environment details (please complete the following information):

• Environment location: Docker rapidsai/rapidsai-core-dev-nightly:22.02-cuda11.5-devel-ubuntu20.04-py3.8
• Linux Distro/Architecture: Ubuntu 20.04 amd64
• GPU Model/Driver: V100-SXM2-16GB, 460.32.03
• CUDA: 11.5
• Method of cuML install: from source, git hash 4e23f87, using tools in the dev docker image