Refine + method

[jennybc]

I merged #297, but then I got to thinking more about recycling. That PR actually changed some behaviour around length-0 input (#286 (comment)).

I think we'll be happier in the long-run to route + through glue_data() rather than paste0() to keep recycling behaviour more in sync. I might add a couple of tests to this PR re: + and input lengths.

[jennybc]

What do you think the correct behaviour is for NULL? 🤔

With CRAN glue (v1.6.2), it's treated like character():

str(glue::as_glue("hello") + NULL)
#>  'glue' chr(0)

With anything after 740ca7f, it's an error:

as_glue("hello") + NULL
#> Error in `+.glue`(as_glue("hello"), NULL): RHS must be a character vector.

The tidyverse recycling rules and #246 don't obviously speak to what to do in the + case.

In arithmetic, it is like adding something of the correct type but of length zero:

1.5 + NULL
#> numeric(0)

[jennybc]

I decided to make glue + NULL return character(). That's was it's been doing and it's consistent with "normal" addition.

[hadley]

I'm not sure that's right; I would have expected it to return glue, because (a) we generally ignore nulls and (b) it makes it more consistent with paste().