Fix query command suggestion in error message with repo mappings

[fmeum]

The package label in the query command suggested when a specified target isn't found in a package looked like @rules_cc~1.0.0//pkg, which doesn't work as the repo name is canonical, but the label isn't.

This is fixed by reusing the logic from Label#getUnambiguousCanonicalForm.

[fmeum]

@meteorcloudy @Wyverald Could you review this?

I noticed that the output of query consists of non-canonical label literals with canonical repo names. Those can't really be used anywhere with Bzlmod enabled. Should query emit canonical label literals instead, at least with Bzlmod?

Edit: I looked into how this could be improved. It doesn't seem that difficult to get query output to use this output format instead: //pkg:file for targets in the main repo, @my_dep//pkg:file for targets in a repo that the main repo has a local name for and @@dep~1.0.0//pkg:file when the repo isn't visible from the main repo. What do you think, is that the right direction?

[Wyverald]

I noticed that the output of query consists of non-canonical label literals with canonical repo names. Those can't really be used anywhere with Bzlmod enabled. Should query emit canonical label literals instead, at least with Bzlmod?

I think to begin with, outputting @@-style labels when Bzlmod is enabled should suffice; but what you described later is even better, if it's not too much trouble to implement.

[Wyverald]

Actually, Yun just pointed out to me that this would break Google, where we don't use external repos at all and @-prefixed labels are forbidden.

Could you instead only turn on this behavior when Bzlmod is enabled?

[fmeum]

Instead of depending on the Bzlmod bit explicitly, I backported the "one label stringification function to rule them all" I came up with for #16483, which decanonicalizes query output as far as possible, to the current PR. I moved the tests over since the current PR doesn't exercise all its functionality yet.

It is slightly weird that the new function takes in a repo mapping of either the main or any other repository, but I found that to be the simplest pattern to gradually improve the existing cases of label formatting. Checking for strict visibility specifically seemed cleaner than testing for Bzlmod specifically, even though the should be (mostly?) equivalent. Looking forward to hearing your thoughts on this!