At home I took dark frames, dark flat frames, flat frames and bias frames. After applying these frames to stacking, the dark bar overlay just did not go away.
I then set to investigate it to understand what went wrong.
In my early collection of master dark frames, the ISO400-30s one shows the most prominent problem. And the dark bar only appears in the "R" channel.
Note that I am not debayering the raw data, simply splitting the RGGB into separate images. And all the images are equalized in GIMP.
I did more tests. Unfortunately the results are inconsistent: under the same ISO, the pattern varied significantly.
Lastly, I tried to see if temperature plays a role here. Many many frames later and after stacking subgroups of frames, I arrived at this picture -- all the master frames (as sub-pictures here) are equalized together so that pixel values can be compared directly.
Evidently, temperature is indeed a key factor here. The bad dark band in R channel becomes visible after the temperature exceeds 35c-36c.
Now the question turns out to be: how to control the camera temperature in the warm and humid local weather?