Hmm, I think any copper fill is going to have rounded corners, given how they are generated.
You can reduce the min-width of the fill area, which sharpens the corners, - not sure what that does to the Gerber file sizes, but I’m sure you can experiment there between Size : Sharpness
Yah, one of the things I’m doing is to reduce minimum width to minimum, and hope it’s less than the fab’s resolution. Another solution is to run a ground trace at the desired distance from the signal trace, and have the interconnect on in. Both options are not perfect.
Is it possible to completely turn off corner rounding for a copper fill zone?
No, because KiCad plots a polygon, which has a finite line width, and then uses the Gerber fill command.
That means the Gerber size will not explode, and you should be able to define widths < 0.1mm, which should way exceed any RF issues.
As a sanity check, PCB is 150ps/inch, so even 1mm is ~ 6ps
How many 10’s GHz are you running here ?
Well, the frequencies I do can possibly go up to 60ghz. However, what you wrote is not srtictly accurate, even at much lower frequencies. For thin t-line separation such as found in 4 layer pcbs 0.1 mm notch is as big as ground separation. True, it’s still much smaller than wavelength, and therefore not expected to cause too much trouble for return loss, but can be undesired when you try to make matching duplicate channels and minimize leakage, and is a bad practice.
Furthermore, multiple notches can be thought of as effective roughness of the ground path which manifest as additional loss, even thought this is not the scenario that is discussed here
Usually I have a t-line segment with high priority level ground plane @ the required distance, inside a board which has ground plane with priority level of 0 all over.