However, you could also exploit certain known characteristics of the image as well as the noise to produce an arguably better image, such as by discarding outlier samples or by using a weighted average based on some expectation of the pixel value.

Yes, that is one use case for pixel binning. Apple uses it to reduce noise in low light photos, but it can also be used to improve telephoto images where more data (from neighboring pixels) can be used to yield cleaner results.