Improve the print method

[hadley]

Inspired by stringr::str_view()

Fixes #290

[jennybc]

This feels like a pretty big change in the print method to me! So much so that it gives me pause.

I've always assumed that the main benefit of having the "glue" class was to get the quote-free, clutter-free printing and this walks away from that. I know we have some regrets about the "glue" class, but this seems to undermine its main rationale?

This almost feels more like a str() method. What do you think of me turning it into that?

(And, if so, yes I still think we need to do something when printing a glue object of length 0.)

[jennybc]

I ran revdep checks and, as expected, only a saw a couple of failures due to this, including r-lib/testthat#1868. But in the course of figuring out that puzzle, I checked pillar locally and saw massive snapshot test change, which I think could be pretty widespread. I know that won't make tests fail on CRAN, in general, but it increases my hunch that we should introduce this display as a str() method.

[hadley]

We should maybe talk this through live as it's definitely not a str method. I think it stays true to the intent of glue to just "show the strings", while avoiding confusion in cases like this:

glue::glue("{x}", x = c("a", "b", "c\nd"))
#> a
#> b
#> c
#> d

^{Created on 2023-09-25 with reprex v2.0.2}

[hadley]

I've resolved the remaining discussion items; @jennybc we should chat to confirm that this makes sense.

[jennybc]

I'm still not sure I'm on board with this, but we'll discuss in person. I'm comfortable with how it's done.

[jennybc]

Do we want this to affect the return value? I.e. print() no longer passes x through invisibly.

[hadley]

See discussion in #290