Tensor storage domain after replication

[DmitryLyakh]

D = Q7 * Q8, where
Q7 and Q8 reside on {0,1},
D resides on 0.
Process 0 replicates Q7 and Q8 from process 1, but D is local.
Current storage domain rule: Tensor operand residence domains must coincide with the execution domain.
Corrected storage domain rule: (1) The execution domain must have access to complete tensor operands; (2) The residence domain of the output tensors must not exceed the execution domain.