There is something fundamentally different in most cases between producing a service that relies on functionality like your location being available, or a camera working, and the use of colour as a requirement for communication.
It may be that we can do something different. Telling authors that the only way we will help them provide an equal experience to people is by breaking privacy on things that are important seems worth trying to avoid, even if it takes some harder thinking.
Telling people that “unless they are prepared to explicitly note they are different, we don’t expect authors to bother catering for their relatively common need” makes me uneasy.
I understand that colour plays a valuable role in communication, and doing without it can seriously reduce functionality and expressivity. On the other hand I am also nervous about dividing the world along lines of colour perception - in part because the reality isn’t very clear-cut, when we deal with devices whose ability to render in the environment they are used varies more than actual people do.
Beyond the issue of revealing personal information, one of the things that makes me uneasy is that we will repeat what I consider the major failing of ARIA, which is that it really only works for the small set of people who have accessibility requirements and happen to use a screen reader to help meet those requirements, while somehow the message we have given developers at large is “ARIA pretty much solves accessibility”. Deuteranopia is relatively straightforward - a continuum of degradation in the ability to distinguish blue from white is far less tractable with media queries. (It isn’t intractable - we already deal with breakpoints in sizes, and device sizes are now so varied that those are chosen as arbitrary values in sensible bands. We can do the same here…).
To return to @fyrd’s comments, the people likely to engage with, and benefit from this should be considered. I suspect an unfortunately large proportion of the relevant developers will be motivated by a requirement to comply with some rule - a situation which encourages the tendency to make what I called “the ARIA mistake” above. Paving that cowpath seems like an anti-pattern to me.
Sorry this comes across as really negative. I think it is worth exploring this space. I suffer regularly from design decisions about colour, based on a fairly minor need for higher-than-average contrast. (The most recent example is in writing this). And I have seen a lot of mistakes in proposing solutions whose unintended consequences are worse than the problem they aim to solve - some of them things that I proposed. Obviously that means I am not infallibly clever (no surprise to those who know me), but sometimes I might have learned something painful and useful…
I note that a number of the technical issues that come up are ones of granularity - something we are also grappling with in the stop-start work on accessibility within schema.org …