`total.03-fe.04-attach_features.11-user-item volume features` stage of HM is 3x slower on MPI than on Ray

[YarShev]

We should figure out why MPI is slow here and fix it.

https://github.com/intel-ai/timedf_benchmarks/blob/b092ea0d490eb630224fc4ffdbc2f62630f57e49/timedf_benchmarks/hm_fashion_recs/fe.py#L159