Validate if mapped memory should always have flush and the workflow in which it should be done

[axsaucedo]

Currently the way that memory is mapped into host memory is always flushed. This in theory is only necessary if the host bit is not coherent, so should ensure this doesn't have a performance issue, or alternatively make sure it's only carried out if the memory has to be flushed (ie if the memory host bit is not coherent).

ie https://github.com/axsaucedo/vulkan-kompute/blob/af4f429d4df40e3b963bc2e106ba0756e80f1d29/src/Tensor.cpp#L202-L212