fix(desktop): workspace git actions enhancement

[kekexunxun]

The previous PR was closed due to the handling of the branch; a new PR is now being submitted. A fallback mechanism has been implemented for the Git "unstage" operation, and code related to a horizontal scrolling overflow issue—which did not require a fix—has been removed.

#3822

[SivanCola]

Thanks for the update here. I found two workspace-scoping issues that should be fixed before this lands:

1. WorkspaceGitCommit currently runs git commit -m without a pathspec. If the user opens a repository subdirectory as the workspace, the UI only shows staged files inside that subdirectory, but Git still commits the entire repository index. I reproduced this with git -C sub commit, which committed both sub/a.txt and an already-staged other/b.txt. Please scope the commit to the active workspace, for example with -- . or an explicit repo-root-relative workspace pathspec.

2. WorkspaceGitStageAll and WorkspaceGitUnstageAll have the same subdirectory leak: git -C sub add -A and git -C sub reset HEAD -- affect changes outside sub/. These bulk actions should also use the same workspace-scoped pathspec, including the unborn-HEAD fallback path.

The Go tests passed locally, and the frontend build passed after regenerating the ignored Wails bindings. A regression test that opens/uses a repo subdirectory would be a good guard for both cases.

[kekexunxun]

Fixes an issue where Git batch operations "leaked" outside the workspace when the workspace was set to a repository subdirectory.

Issue

When a user opens a repository subdirectory (e.g., sub/) as the workspace:

• git commit -m (without a pathspec) commits changes from the entire index, rather than just files within the workspace.
• git add -A and git reset HEAD behave similarly, affecting files outside the subdirectory.

Fix

Added the -- . pathspec to three batch operations to restrict them to the workspace directory.

Testing

Added TestWorkspaceGitScopesToSubdirectory to verify that StageAll, UnstageAll, and Commit do not leak into the other/ directory when sub/ is the workspace, and to cover the rm --cached fallback path for an unborn HEAD. All relevant tests passed.