I think one reason is that image resizing algorithms are nontrivial. An image uploader sounds simple enough until you start supporting many different scaling algorithms [1] across a bunch of formats. Browsers are already big and complex enough IMO. This could be easily / best handled by a library instead.

Another point is that, holding onto the unmodified images server-side (and merely serving a cached, resized, optimized version on the fly) can be beneficial for future use cases.

[1]: https://en.wikipedia.org/wiki/Image_scaling