The point of the exercise being that you seem to be dismissing the fact that there are actual bad consequences to the 'must be 100% accessible upon first release' principle, simply because a person for whom accommodation is required now has a job.
What I don't see in your argument is some sort of equitable discussion of who bears what costs, and how that equity is fairly determined.
> lose their job because it now requires an app that they can't use.
So is it a convenience or essential for the job? If it is a convenience, then it isn't essential. If it is essential, you're denying it to the larger group in the interim.
Maybe we should just go ahead and gouge everyone's eyes out to really level the playing field...
Keep in mind that things take time to develop a userbase, and that large companies really like things to have an established userbase and support before they get on board. For instance, one of the major reasons that the Raspberry Pi Foundation has seen so much success, even though they aren't the fastest or the cheapest, is because they have a large community support network and support their stuff well.
As such, as long as I (and / or management) are committed to implementing solely-accessibility related features in the near future, I would take the first option, and release as soon as I could.