Is the long-term goal for tidy evaluation that it will remain a special feature of the tidyverse family of programs, that it will ultimately be extended into base R functions with nonstandard evaluation, or that the tidyverse itself will expand to offer tidy-eval alternatives to most important or commonly-used base R functions?
If I understand your question correctly, there was a recent thread which you might find useful and interesting.
Yes, I think that thread eliminates the second of my three alternatives.
The difference between 1) and 3) is whether the tidyverse aspires to becoming the primary mode through which users interact with R, at least during the data preparation and exploratory analysis stages.
Suppose you are writing a package. At this point, using, e.g. dplyr in your package code entail a larger maintenance cost because dplyr is still rapidly evolving. But that cost may be partly, fully, or more than fully offset by the benefit of familiarity as the tidyverse becomes a more "standard" method of interacting.