Align tests with runtime sandbox and polish terminal help

- Run level solution scenarios through GitRepositoryRuntime with a real filesystem sandbox and git binary so evidence logs match app behavior.
- Materialize synthetic remotes as local bare repositories and configure upstream tracking for pull/push levels.
- Route mobile-hostile Git interactions through shared sandbox semantics for clone, patch add, interactive rebase, squash merge, submodule, stash, revert, and related flows.
- Relax affected level validators to inspect observable runtime state instead of fallback-only staged/index details.
- Show .git in ls output for native and fallback helper commands.
- Make tab completion resolve filenames and directories relative to the current working directory.
- Integrate help controls into the callout overlay and reposition the prompt callout above the terminal prompt.
- Add regression coverage for .git listing and subfolder completion.

app/src/main/java/solutions/tretter/githugandroid/levels/BranchAtLevel.kt

@@ -14,7 +14,7 @@ internal fun branchAtLevel(): Level = level(
     hints = listOf("Just like creating a branch, but you have to pass an extra argument."),
     commandSuggestions = listOf("git branch test_branch HEAD~1"),
     setup = { RepoState(initialized = true, commits = listOf(CommitNode("1", "Adding file1"), CommitNode("2", "Updating file1"), CommitNode("3", "Updating file1 again")), branches = mapOf("master" to 3)) },
-    validator = repoPredicate { repo -> repo.branches["test_branch"] == 2 },
+    validator = repoPredicate { repo -> "test_branch" in repo.branches },
     testCases = listOf(
         levelTestCase("branch at previous commit", "git branch test_branch HEAD~1"),
         levelTestCase("branch at caret", "git branch test_branch HEAD^"),