Document why GitProcess does not use Enlistment.DotGitRoot

[tyrielv]

Summary

In a linked git worktree, Enlistment.DotGitRoot and the git directory that GitProcess passes as --git-dir point at different locations:

Field	Value in a linked worktree
Enlistment.DotGitRoot	the shared .git directory of the main repo
GitProcess git dir	<worktree>\.git — a file holding gitdir: <shared>/.git/worktrees/<name>

This looks like an inconsistency. It is not. Each value is correct for its own purpose, and the divergence is load-bearing.

Git follows the gitdir: pointer, so the worktree .git file resolves the per-worktree state (HEAD, index, per-worktree refs, reflog) and, through commondir, the shared state (config, objects, packed-refs). DotGitRoot is used for shared state that VFS for Git reads or writes directly on disk (objects, objects/info/alternates, hooks).

Why "aligning" the two would be a regression

Changing GitProcess to read Enlistment.DotGitRoot binds every command to the main worktree. Verified against a real worktree with divergent HEADs (git 2.55):

--git-dir	rev-parse --abbrev-ref HEAD	log -1 --format=%s HEAD
<worktree>\.git (current behavior)	feature ✅	onfeature ✅
<shared>\.git (DotGitRoot)	main ❌	onmaster ❌

So from inside a linked worktree, rev-parse, name-rev, branch, symbolic-ref, update-ref, and read-tree would all silently operate on the wrong worktree.

The second GitProcess(gitBinPath, workingDirectoryRoot) constructor needs no change — it has no Enlistment to consult, and its computed path is already the correct one.

What this PR changes

Nothing functional. It documents the divergence and pins it with tests so a future "cleanup" cannot quietly introduce the regression above.

• GitProcess.cs — comment explaining why the git dir is deliberately not Enlistment.DotGitRoot; new GitDirectoryPath accessor so the value is testable.
• GVFSEnlistment.cs — comment recording that DotGitRoot is shared-state-only, and that per-worktree paths must come from Worktree.WorktreeGitDir (as GitIndexPath already does).
• WorktreeEnlistmentTests.cs — two tests pinning the current values.

Testing

• Full unit suite: 919 passed, 0 failed, 11 skipped — identical before and after.
• Mutation-verified: applying the rejected change (GitProcess reading Enlistment.DotGitRoot) fails GitProcessUsesWorktreeGitFileNotSharedGitDir and nothing else, confirming the new test actually pins the behavior.

Note for reviewers

The audit behind this PR also found callers that use DotGitRoot for genuinely per-worktree state, where GitIndexPath already resolves correctly — FileSystemCallbacks.cs (watches the main repo's logs/HEAD), GitIndexGenerator.cs, RepairJobs/GitIndexRepairJob.cs, DehydrateVerb.cs, and FastFetch/CheckoutPrefetcher.cs. Those are real divergences with real blast radius, and are deliberately left out of this PR to keep it a no-op. They will be evaluated separately.

Targets vnext: long-standing behavior, not a 2.0 regression.