Enable `PreMulSum` User-Defined Reduction Across All Reduce-Type Collectives and Scale-Up/Scale-Out Configurations

[pkourdis]

Summary

The following gaps have been identified for PreMulSum reduction:

• Missing PreMulSum user-defined reduction in AllReduce for scale out
• Missing PreMulSUm user-defined reduction for other reduce-type collectives for scale-up and scale-out (internal MLSL-4016)

Justification

Support XCCL backend in TorchComms.