In camera calibration, fx and fy are tied to the image coordinate system: the spacing and scaling of pixels in the recorded image, not the exact light-sensitive shape inside each sensor site.

So if the output image uses square pixels, calibration would normally still give fx ≈ fy unless there is some real anisotropic scaling in the imaging chain. The physical focal length of the lens itself also does not depend on sensor pixel or photodiode shape.

What photodiode fill factor or shape can affect is sensitivity, noise, or how a manufacturer reconstructs the final image. But masked, split, or partially covered photosites do not by themselves change the calibrated focal length terms, because calibration is performed on the final sampled image grid. The same idea applies even to scanned film: calibration works on the pixelized image representation.

In short: calibration depends on the geometry of the final image pixels and optics, not the detailed shape of each photodiode.