simplify and improve printing of qualified names

[JeffBezanson]

This removes repeated code, fixes some cases where quoting was not done
correcly (e.g. Mod.:+), and checks identifier visibility recursively
so only a minimal module path is printed.

fixes #39834

[JeffBezanson]

Ah, so Documenter uses a temporary module with an invalid identifier name plus some regex magic, which is causing the weird Main.var"Main" printing. Our options:

• Don't use var"" here, rolling back part of the fix. That would be bad because we were printing these incorrectly before.
• Add var"..." to the doctest heuristics.
• Use a valid identifier for the doctest sandbox name

[JeffBezanson]

Of course this ended up being much more involved than you'd think. The main issue seems to be whether to do this for print/string as well as show/repr. So far I kept the output of print(::Module) the same (always fully qualified) to try to be less breaking. However, that still breaks some tests similar to

@test sprint(show, x) == "$(@__MODULE__).A.B.x"

since interpolation uses print. So the questions are

1. Should we do the new auto-qualification for both print and show?
2. Should we address RFC: make default module-prefix showing independent of the state of Main(e) #29466 at the same time?

If print and show are different, that can be confusing of course. But if they're the same, we hit #29466 in that the result of string(x) or interpolation can change based on using things at the prompt. But that trades off against lots of package tests relying on the Main behavior --- the test file does using MyPackage, so it's visible in Main, so things will print un-qualified. Maybe we could add some feature to temporarily change the global default module setting for printing?

[JeffBezanson]

Triage says: this is maybe justifiable if we combine it with something else that makes it easier to write these kinds of tests. Maybe combining it with #29466 would do that, by making repr(x) by itself more predictable.