ability to perform runtime autotuning

[slaizet]

ability to perform runtime autotuning of the process grid dimensions used to partition the global domain and communication backends used for transpose and/or halo communication. This feature enables users to run the library using the best performing configuration for a given global domain size, number of tasks, and compute cluster topology. The autotuner aims to select decomposition and communication backend options that minimize transpose and halo communication time

See https://nvidia.github.io/cuDecomp/autotuning.html