However the whole point of TPI is to reduce the flow temperature of the water when you are near the desired room temperature [...]
Completely wrong. The whole point of TPI is that it has to be used when such direct proportional control is unavailable. As you so eloquently put it: 'it switches the boiler on and off'. It us unable to lower the flow temperature directly and so can only vary the on/off duty cycle to lower the overall, average, heat output. Flow temperature control remains solely in the realm of the boiler.
The OP's question is very valid - in the described scenario we have a controller attempting to effectively reduce the heat output using a variation of pulse-width modulation i.e. lowering the average output through discrete on/off duty cycles however we also have a boiler attempting also to reduce the output based on its own assessment of demand (indirectly through measurement of flow and/or return temperatures). In the absence of any direct linkage between the two I think the OP has a very valid question and one which I am unsure of the answer to, particularly when you add into the mix the behavior-learning properties of most of these TPI stats as whilst it sounds like a grandiose capability they seem to be doing little more than recording the long term rate of rise/fall of the system as a whole which will of course vary in the short term if the boiler starts lowering its output.
The solution of course is simple, albeit no good for existing systems - choose a boiler that allows the external stat to directly control the heat output and modulation of the boiler hence it is in control of every aspect.
Mathew