Chemistry is definitely not my area of expertise, however; some further thoughts:-
In the conventional diesel engine, although the global air/fuel ratio is lean, the local air/fuel ratio on the periphery of the fuel plume in the region of the flame, is close to stoichiometry and that is true whether you are at light load low idle or at full power rated speed. For acceptable fuel consumption (sfc), the full load injection duration must be kept under about 30 crank degrees, but the heat release can extend much longer than that.
Now, the flame is virtually stationary since it is anchored by the fuel plume, even in swirl supported combustion systems. So you have a near constant heat source with a flame that is in excess of the ~1900°K thermal NOx formation temperature, and this is surrounded by plenty of extra air that is wafting over and feeding the flame. My assumption had always been that the NO and NO2 was formed from thermal breakdown of the nitrogen in the air rather than the oxygen others have mentioned, and since there is plenty of that in a diesel engine even at full load (since the fuel plume is not able to reach all the air in the chamber), we get NOx. This is still true but less so, even when we use a diluent such as EGR.
In the homogeneous stoichiometric SI engine though, the fuel burns in a flame front that radiates out from the spark plug. In this scenario and assuming good homogeneity, the flame may well be over the thermal NOx threshold as it is in the diesel engine but it fully consumes the air behind the flame front and progresses into the heated air in front of it perhaps limiting its opportunity to form NOx? There is also a time constant in the formation of NOx.
In the case of the true HCCI engine, the air/fuel ratio is both homogeneous and way lean of the flammable range so combustion is a near simultaneous auto-ignition event distributed all over the chamber, with no flame front at all. As such, the AFT remains below the thermal NOx threshold.
PJGD