CLEANUP: globalsOf() should not return `...`, if `...` is not a global

Adopted from #72: Currently `globalsOf()` reports on `...` even when it "already knows" it's not one, e.g.

```r
expr <- quote(fcn <- function() ~ list(...))
print(expr)
# fcn <- function() ~list(...)

names <- globals::findGlobals(expr)
print(names)
## [1] "<-" "~" "list" "..."

globals <- globals::globalsOf(expr)
str(globals)
## List of 4
## $ <- :.Primitive("<-")
## $ ~ :.Primitive("~")
## $ list:function (...)
## $ ... : 'DotDotDotList' logi NA
## - attr(*, "class")= chr [1:2] "Globals" "list"
## - attr(*, "where")=List of 4
## ..$ <- :
## ..$ ~ :
## ..$ list:
## ..$ ... : NULL
```

Note that `structure(NA, class = "DotDotDotList")` global object with a `where` attribute being `NULL`. This is meant to signal a _potential_ misuse. However, I think it's time to graduate **globals** from this conservative approach and instead drop that `...` element, i.e.

```r
str(globals)
## List of 3
## $ <- :.Primitive("<-")
## $ ~ :.Primitive("~")
## $ list:function (...)
## - attr(*, "class")= chr [1:2] "Globals" "list"
## - attr(*, "where")=List of 3
## ..$ <- :
## ..$ ~ :
## ..$ list:
```

Comment: To do this, we also need to adjust the current package tests that assumes `...` is returned.

BTW, can we fix this also for `globals::findGlobals()`?

Start by locating the implementations of globalsOf() and findGlobals(), then inspect the package tests that cover the `...` result. Reproduce the example in the issue and update the behavior so globalsOf() omits `...` when it is not a global. Confirm the relevant tests reflect the three-item result and determine whether findGlobals() should change as well.

Written by the indexing model from the issue text.