A radio signal recording of a wireless communication system (e.g: Wi-Fi traffic) is beaconized, channelized and subject to noise.
When working with such an RF signal, numerically transformed to a stochastic timeseries, to which extent can I consider the signal to be stationary?
That depends on your signal model. We can't tell you that – but for example, in time-slotted system, obviously the received signal can't be stationary – its variance (which is basically its power) depends on the time you're looking at.
We tend to model noise as (weak-sense) stationary; same for data coming out of the source coding; we often even technically enforce that using whitening scramblers.
However, that really depends: some systems use whitening. Others don't.
Some techniques assume all disturbances to be stationary. Others are built around the specific notion that interference is not stationary.
So, model your signal.
Any pointer papers about signal stationarity in telecommunications are appreciated.
Not an issue for papers: this is subject for standard textbook discussion; Goldsmith and Viswanath/Tse would be my go-to books here.
(Generally, if you feel like something is so important that it is a basis for all that you do, it's almost certain people have looked at it a long time ago and have already written a good book on it. The trick is finding the good books; in this case, things are so fundamental that basically all wireless communication books concern themselves with it.)
What properties are relevant from a communications point of view?
All and none. Defined in the signal model.
