Smarter Lustre OST mapping

The current way of dealing with Lustre striping when Snapshots:lustre_OST_count is set does not perform optimally when the number of OSTs on the file system (N_OST) is larger than the number of ranks (N_rank) or when N_rank%N_OST > 0, since it is biased towards low OST indices. The situation is especially bad when you have e.g. N_rank=48 and N_OST=192, since then only the first 48 OSTs are actually used.

It would be better if the decision on which OSTs to use could be out-sourced to a function that introduces some more randomisation to make the OST usage less biased.