I know exactly what you are talking about. Generally I have the router bit out of the wood before starting to run gcode to avoid that, but if the router bit is in contact with the wood when the program starts it will jerk like that as it wakes up. Are you creating these squares using the jog command buttons?
Yes, in fact it might already be an option in FluidNC. It looks like this is what we want: Axes | Wiki.js
Unfortunately it seems like the maximum value that they support is 250ms or it can be disabled entirely. It shouldn’t be too hard for us to change that option to be longer if we would like.
My thinking exactly 
My thoughts generally on accuracy is that it was the next thing on my todo list a couple weeks ago and then I got sucked into fighting fires. It’s the top on my todo list again when I have a chance.
Here is my go-to testing file (three 600mm squares, just the corners):
SD_Test Squares.nc (6.9 KB)
There are a couple of known issues right now that are worth being aware of.
1st - The machine doesn’t have a good way to home the z-axis right now. When it powers up it believes that the spacing between the top of the z-axis motor and the bottom of the lower router clamp is 28mm regardless of if it actually is. The only way to get that right now is to lower it all the way down…raise it back up 28mm…power off…power back on which is a huge pain. I was working on making a button to do that automatically, but integrating it into FluidNC was turning out to be a little tricky. This is top of my ToDo list.
2nd - I had some code in there which was scaling everything by like 105%…basically my tests were coming out 5% smaller in both X and Y than expected so I scaled everything by 105% and it was good…and then I forgot about it. I came across that code later and it was such a hack that I took it out again so I wouldn’t forget. I think that what is actually going on is that the tooth spacing on the belts is not exactly 2.00000mm it’s like 1.9999mm or 2.0001 mm and we need to do some testing and put the right number in for that. I think that the only way to get that dialed in is to put in a new number…run calibration…cut test squares and measure them…tweak the number…repeat. At least that was my plan.