Of course any bitmap image can be vectorized, but might we ever reach a point where a camera can make an outline image that looks exactly like a photograph? Could the resulting image even be considered a photograph, and if not, what defines a photograph?
I think the problem here is one of level of detail. While vectorization works great for making an image that can scale infinitely, it relies on firm patterns that can be mathematically described.
Unfortunately, real life is full of imperfections and variations that make it impossible to describe in a pure vector format, at-least with any meaningful gain. We could theoretically make a vector format that maps every single pixel, but then we'd have a raster image that couldn't scale any better than a normal raster image.
Image and video compression are already the applications of the kind of thinking you are talking about. They look for the patterns that can be identified to reduce the storage required and when using lossy compression, they bend the rules further to get a match so that they can reduce the amount of information necessary to represent the image.
Vectorizing the image is another level of extreme for this, but you will notice that the image quality always drops significantly when such vectorization is applied (due to the loss of the random information that makes a photo look like real life.)
As for whether a vector image of a scene is a photo or not. I think that's a really hard question to answer. Personally, I'd say yes, if it is a rendering of real life based on sampling the light (regardless of how lifelike it may be), I think that it probably could be considered a photograph, but I could also see how someone might see it as more akin to a painting at that point as well. I don't think there is a strong line answer there.
Such cameras exist today such as those used to survey road accident sites.
Those cameras create a 'point cloud' of vector distance measurements based on reflected laser light and there is no need to be evenly spaced or co-located pixels although they normally are (as it's easier to do it that way). There are ways which that captured image can then be displayed using interpolated projections (including on vector displays as used to be used by CAD operators in days gone by) or by printing using either material removal (like CNC) or an additive process (such as Selective Laser Sintering.) There's no reason why such a camera couldn't record colour information and they may even do that anyway.
The word Photograph is a compound of Photo (light) and Graph (record) as such, yes it would be photography as it would be recorded light.
Please humor me:. Consider that "randomness" in nature is mostly not true randomness but fractal, stochastic and other complex but all formal mathematical patterns. Imagine AI that can accurately extrapolate the patterns from an image with a high degree of certainty --- definitely including all the data in a raster image of whatever resolution, but also including the infinitely complex patterns between the pixels. Isn't this possible?