In general, when developing a resistance (or powering) prediction for a design using CFD, it is important to account for any factors that may not be accounted for in the CFD simulation.
a. Things like Ca are strictly speaking a model-ship correlation factor. So if you run a CFD simulation at full-scale you might argue that it should not be necessary to include, however, some Ca formulations include a estimate of hull roughness so even if you choose not to include Ca you should include some factor to account for the fact that the CFD simulation is for a perfectly smooth hull whereas most commercial vessels will have some drag due to roughness, even if it is new anti-fouling paint.
b. Other additions really depend on what was included in the CFD model. Early on before the superstructure/pilot house are well-defined it is common to run simulations of just the hull. In those cases it can be important to add in an engineering estimate of "still-air drag" for the missing superstructure.
c. If you run a bare hull model, you will need to remember to add drag for appendages like rudders, shafts, struts, etc.
d. Some naval architects/design firms have a policy of adding on a % margin to account for those items that may not be included in the CFD. What that value should be is often based on past experience with those particular types of designs. Sometimes a margin is added on for other reasons, such as to account for uncertainty in the prediction, or uncertainty in the design which may change over time. For example, some organizations have a resistance and powering margin policy where the magnitude of the margin depends on the stage of the design. Early stage designs prior to having a high confidence even in the shape of the design may use a 10% margin for example while a lower margin, say 5%, is applied during the later design stages as the propeller and other details become more clear.
As for uncertainty in the CFD predictions themselves, much has been written about trying to quantify CFD uncertainty. For example see the ITTC document, Uncertainty Analysis in CFD Verification and Validation, Methodology and Procedures. Our experience with Orca3D Marine CFD when comparing to experimental and/or full-scale data has been that the predictions of resistance generally agree within about 5% assuming you have ensured that the CFD simulation is converged (both in space and time, i.e. meshing and time step).
Ideally the CFD model is a complete model with all relevant geometry included, but sometimes that is not the case. The Orca3D reporting tool gives you the flexibility to add resistance components to account for things that were not in the CFD simulation.