Between writing a single line that includes a 100 MB library and writing a hundred lines that include just the code needed; many people will pick the former. Just look at usage of GPUImage or other libraries.
Certainly not the ones at Facebook, Google, etc. Nobody gets promoted for “we shipped less new features but our latency remained constant”, or “we shipped less new features but reduced device battery use by 10%”.
I doubt the total installed size of the FB app is very high on their list of concerns.
Key word right there. One criterion you didn't explicitly list: development time.
Pruning and deduplicating dependencies takes time and effort that could be spent elsewhere - e.g. all the criteria you listed - and so the question becomes, what benefit is there to improving that? There's an argument to be made for onboarding, of course, but between the prevalence of WiFi and the proportion of users with the app installed to those installing it, I suspect it's tough to make.
(For lots of pictures and videos, you can keep a lower quality version on the device as a compromise---but not sure whether Google Photos or Apple does that by default?)
The problem isn't so much the speed but the caps, if you're not on Wifi. I don't know anyone who has more than 5GB/month, and HD video will eat that in less than an hour.
The UK isn't ahead of the rest of the world here, is it? (Google paid for my mobile data in the last few years before that, so I am out on the loop on general prices.)
Any decent product team – which I assume these apps have – is tracking this, and is continuously setting direction for the tradeoffs among space optimization, speed optimization, bandwidth optimization, battery life, usability, stability, server costs, feature velocity, and probably a half dozen other factors.
The implication of some of the comments here is that they've been making these tradeoffs incorrectly. I'd be interested to hear more about the reasoning behind this conclusion.