It IS a NEW sensor. Canon reports it consumes less power than the old 18MP sensor (not to mention the hybrid AF). It is NEW.
"The same number of pixels in the EOS Kiss X5, has developed a new image quality and also to enhance performance. Support high-speed readout of 5 fps / up to about, and has realized the imaging plane phase difference AF (hybrid CMOS AF), and further power saving."
I don't think there's doubt that it's 'new' but rather whether there is any actual difference in the aspects of the sensor which record the captured image. Similar to my reference above, the EF-S 18-55mm f/3.5-5.6 IS II is clearly a new lens relative to the original, but it's optically identical, meaning from an image forming standpoint, new ≠ different/better.
You have to be careful reading Canon's marketing materials when they use words like 'new image quality' (especially when it's a Google translation - recall that the original translation of the spec list had the 650D with a pentaprism, when it's a pentamirror like all the other Rebels). Canon throws out terms like 'two full stops better' without specifying that it's almost all in the RAW conversion.
The changes you describe could easily have nothing to do with RAW image IQ.
Besides, Digic 5 is VASTLY more important than pixel count, vastly.
Vastly important IF you shoot in jpg, yes. If you shoot in RAW, then Digic 5 is relatively unimportant as it relates to image quality.
I am curious, how can they build (layer) a phase detection sensor over an existing image sensor without causing filtering effects on the actual image (unless they use some algorithm to actively fill in those empty/ light blocked spots)
Even with 'gapless microlenses' there are spaces between photosites, which are at the back of the sensor in any case. PDAF sensors are line sensors that can be made pretty thin.
Also, as you state, they could easily interpolate around those small areas - consider the semi-automatic remapping of dead/hot pixels when you manually clean the sensor.