IsClonedAsync() returns true for an empty directory inside another repository's working tree, so "clone if not cloned" skips the clone and later verbs run against the outer repo #185

What's wrong

GitRepository.IsClonedAsync() is documented as deciding "whether LocalPath currently holds a git working tree", but it delegates to GitProbes.IsWorkTreeAsync, which runs git rev-parse --is-inside-work-tree:

• GitIntegration/GitRepository.cs:93-111 (at 647dbbc)
• GitIntegration/Execution/GitProbes.cs:37

GitResult<string> result = await new GitTextBuilder(runner, path, "rev-parse", "--is-inside-work-tree")
	.TryExecuteAsync(cancellationToken).ConfigureAwait(false);
return result.Success && string.Equals(result.Value, "true", StringComparison.Ordinal);

--is-inside-work-tree walks up the directory tree, so it answers true for any directory under an ancestor working tree, including an empty one:

$ git init outer && mkdir outer/sub
$ git -C outer/sub rev-parse --is-inside-work-tree
true

Failure scenario

A launcher enumerates repositories, maps each to a GitRepository with LocalPath = <root>/<RepoName>, and follows the "if not cloned, clone" pattern. If <root> is itself inside a git working tree (for example a dev folder that is a repo, or a monorepo checkout), then:

1. IsClonedAsync() returns true for the empty <root>/SomeRepo directory, so the clone is skipped. A manual Clone(...) into the same path still succeeds, which shows the path was never a repository.
2. Status(), Log(), Commit() and the other verbs on that GitRepository then run against the outer repository, with no error.

The reviewer reproduced this against a build of HEAD with git 2.43: IsClonedAsync printed True for an empty directory inside another repo, and cloning into that directory then succeeded.

The repo has already fixed this for Init(). GitInitBuilder.cs:~133 notes that --is-inside-work-tree "wrongly reports AlreadyExisted = true for a plain subdirectory of an existing repository", and #138 / c246cc6 added GitInitBuilder.IsRepositoryRoot. IsClonedAsync still has the upward-walk problem.

Suggested fix

• Make IsClonedAsync answer "is LocalPath the root of a working tree". Either:
  • require rev-parse --show-cdup to exit 0 with empty output, or
  • reuse the --is-bare-repository --git-dir --show-cdup probe and GitInitBuilder.IsRepositoryRoot.
• Leave IGitClient.IsRepositoryAsync / OpenAsync on the "inside" semantics, since their contract really is "inside a working tree".

Acceptance criteria

• IsClonedAsync() returns false for an existing directory, empty or not, that is a subdirectory of another repository's working tree.
• It still returns true for a clone root, including a submodule, a linked worktree, and a --separate-git-dir checkout. These are the cases Init() reports AlreadyExisted = false when re-initializing a submodule, worktree or --separate-git-dir repository, hiding an ignored --initial-branch #138 had to handle.
• Integration tests cover the nested-directory case.

Related, but a different symptom: #178 (verbs misbehave when LocalPath is a subdirectory).