The receive side of this device is an I-Q mixer, from the SSB noise figure if you could cancel the unwnated sideband would be -174dBm (kTB in Hz) +10dB NF + 20dB gain. The I-Q bandwidth is not specified, in its intended application would only need to be very high, about 10kHz.
Given the quadrature and amplitude accuracy specification inwanted sideband cancellation is likely to be as good as non-exitant so the noise figure would increase by 3dB.
Without taking the AM and phase noise of the oscillator into account the noise power in a 10kHz bandwith at the IQ output will be about -100dBm, TSS wwould be about -92dBm.
The AM noise will have an effect even if is likely to be around 30dB higher at 10kHz than the 100kHz offset quoted, this will depend on how good the ocillator is it could be 10dB better, but then it could be 10 worse. This noise power per Hz will keep increasing as the offset decreases, but the bandwidth is also decreasing; you will have to integrate the noise power over the band you want to use. I am guessing you would lose at least another 3dB of sensitivty from local oscillator AM noise.
The above are 'back of envelope' calculations. A fuller ananlysis would take more time than I have to lok at it, but hopefully gives ou some pointers
Peter