In the interest of having an answer and since this is CW anyway:
English is not my native language and (thus) I read several articles/books/chapeters on mathematical writing in English, among others by Halmos, Krantz, and Knuth (et al.)
I do not have the references handy and, since it's been a while, do not recall what precisely they said (and in particular not who said what exactly), but the consesus is definitely 'we' (with exceptions for things like thanks and so on; or, somebody said roughly: one of the rare situations where 'I' feels appropriate is to say 'I could not prove this conjecture' where both 'we' and 'one' seem odd; and also for the 'thanks')
So, I strongly recommend using 'we', in the sense you mentioned yourself (the author and the reader) and using 'I' only if it is a really somehow personal statement.
There is the additional question of using 'the author' instead of 'I' (in the above exceptional cases). Here, I think the situation is slightly less clear. Typically, if I am the sole author I write 'I' as writing myself 'the author' feels somehow strange for me. But, I had at least once a journal that changed this (in the Acknowledgment something like 'I thank..' was change to 'The author thanks...').
In case of multiple authors I am a bit more likely to write 'the authors' (in the exceptional cases) in order not to have two different 'we' (authors and reader as well as just the authors).
Personally, I think an additional advantage of the 'we' (instead of 'I') is that thus articles are written in the same way no matter whether it is single authored or coauthored.
To have single authored ones with 'I' and coauthored ones with 'we' seems confusing (well, maybe confusing is a too strong word, but I'd find it at least mildly distracting).
And, finally, how incovenient would it be, one starts a project alone and starts to write a bit of it up, and latter one finds a coauthor...then one would have to change all the 'I' to 'we'.