Update verification check to save memory

[tomdeakin]

The OpenMP CPU version allocates twice the memory it needs restricting the maximum problem size.

For offload models, this is OK as the data needs to exists on the device during the run and on the host during verification.

This could be solved by asking the implementations to pass a pointer to host data back to the driver routine for checking, rather than copying into a vector allocated by the driver.
This would allow implementations to do save memory where possible.

[tomdeakin]

Refers to this: https://github.com/UoB-HPC/BabelStream/blob/7d570547edf0f0222d022d8cd1d67f768e666738/src/main.cpp#L330C3-L330C3