Dunno if its feasible.
But take into account that this would result into a matrix of 336000 x 336000. Depending on how this is done it wont be very… computationally feasible. You could use a downsampled fsaverage space.
This being said… I am not sure if it does make sense tu use each vertex as a node, given that DWI data comes from a lower-resolution and you will have smoothness and, as a result, redundant info there…
Anyway, I am not familiar with tck2connectome nor if it accepts surface images. However, if your goal is to create a dummy/fake parcellation, I would load the surface using python/matlab, and generate a fake parcellation with numbers going from 1-330k and save it as .annot.
This should do the trick.