Make rootFS's showPath() render the paths from the original accessors

[edolstra]

Motivation

This makes paths in error messages behave similar to lazy-trees, e.g. instead of store paths like

         error: attribute 'foobar' missing
         at /nix/store/ddzfiipzqlrh3gnprmqbadnsnrxsmc9i-source/machine/configuration.nix:209:7:
            208|
            209|       pkgs.foobar
               |       ^
            210|     ];

you now get

         error: attribute 'foobar' missing
         at /home/eelco/Misc/eelco-configurations/machine/configuration.nix:209:7:
            208|
            209|       pkgs.foobar
               |       ^
            210|     ];

This should also enable fixing #4507 (similar to lazy-trees) but that will require a bit more work.

Depends on #12512.

Context

---

Add 👍 to pull requests you find important.

The Nix maintainer team uses a GitHub project board to schedule and track reviews.

[dpulls]

🎉 All dependencies have been resolved !

[roberth]

It seems that this would affect the equality of roots. That seems acceptable for entirely flake-based code, but some Nixpkgs and third-party code will be used in this context without the code or their maintainers being aware of the change.
If I'm correct about the following, this needs a release note describing the new phenomenon.

let rootOf = lib.converge dirOf;
in rootOf self.outPath != /.;

This seems to bring us close to the semantics of #11367, which also introduces distinct roots, but I dont think this PR implements something to avoid creating those <hash>-<hash>-source paths yet (which could of course be done later).

[Ericson2314]

I think it would be more future-compat if this did not use CanonPath keys, both for lazy trees and for Windows.

[edolstra]

Right now it must use CanonPath, because we don't have a concept of non-CanonPaths in the virtual filesystem exposed in the evaluator.

[edolstra]

It seems that this would affect the equality of roots.

I don't think so. This PR should only affect the display of error messages. It shouldn't change the semantics of evaluation.

[edolstra]

Closing for #12915.