Because each have diferent reasons (and so can be helped/worsened in different ways).
And the calibration will help only in specific setup and needs be done differently for different setups.
The easiest way to understand is if you take two cameras, whitch are ideal, just not parallel - then you get the image of the same on different parts of resulting image and it would be worse with distance, so you can set it up perfectly for one distance and aproximate that for all other distances.
On the other hand if your cameras are ideal, just their distance is not perfect, you get the same image as in previous case, but this time the solution is to just substract the missalignemnt from the position and the error would became less with distance.
But if your cameras are rotated, you need to transform the whole picture of one of them to align the item on both - again different problem and solution.
The lens distortion can be the same for full image, or can differ in different parts - mainly in corners - so if your object of interest would not be in centre, you have another complicated problem to solve.
So if you are to use it at given angle and distance, you calibration can be a lot simpler, then if you want to use it on wide range of distances and angles to measured object.
And if you want to write your own software/dirvers/addons, you need to know all those sources of errors, as you may need to include them (and their respective corrections) in your software. And it is better to measure them yourself with your piece of HW, than to rely on average tolerance given by manufacturer for the whole production line.
Also the corrections can be done fast (and somewhat inaccurate) with a little space needed, or slow with a lot of sources consumed but giving better precision - even if manufacturer provides some correction method, it is necessarry compromise to somehow fit needs of mayority of perceived customers while not taking too much time and sources.
But you particular application may have different criteria and would like to have at least some results really fast (racing car), or need as precise results as possible (scientific observation), or needs just something with less work needed (hobby project), so it is good to mention, what sources of error are there and how they differs, so you can choose the approach optimal for you.
