My speculation is that the vision cars' software was originally branched from the "mainline" code of Tesla software into its own branch, then vision-only changes are made. Each time the mainline has some changes developed, Tesla either needs to merge those changes into the vision branch, or rebase the vision branch on top of the latest mainline branch (still a merge but in different direction). Either way, there will be code conflicts to be resolved, and extensive testing done to make sure the merge and conflict resolution are good. This extra work takes time.
For the long run, if Tesla indeed commits to vision-only cars, at some point in the future the vision branch will become the mainline, and the radar branch (old mainline) will start to require the extra time to merge and test. Let's see when this will happen.