In general, this is impossible in the simulator. This is a constant phase shift of 45 degrees, and that would be non-causal in the time domain.
It could be done in frequency domain analyses only (e.g. ac and hb, including PSS in hb mode), but not by using Verilog-A. Verilog-A cannot be written in the frequency domain - you'd have to translate it into the time domain, and since there is no time domain representation of such behaviour (except maybe over a narrow band), it can't be done that way.
The way I would do this is to create an s-parameter file representing the complex transfer function, and then use nport with this. Be aware though that simulating it in time domain analyses would either fail or give bad answers, as it is not possible in real life. When creating the s-parameter file, you should make sure there is a dc point which doesn't have a complex transfer function (as that's not possible) to ensure the DC behaviour is correct and it can find an operating point.
There is an internal implementation of a phaseShift for SpectreRF use (in frequency domain analyses), but it's not really in a state that I'd want to post it on the forum.
Why can't you do this in the post-processing tool? That works fine - simply multiple your signal by complex(sqrt(2) sqrt(2)) in the calculator.