Fix CI-Tests - solve false-negative caused by poor setup (issue #703) - #702

Raised in review, and worth recording here:

the key is hardcoded as @"PBShowStageView" rather than shared with
PBGitDefaults. That matches PBGitRepositoryFetchArgumentsTests, so
it's house style, but a rename of the constant wouldn't break the test

• it would silently stop defending.

That is right, and the answer was already in the test this one was
modelled on. PBGitRepositoryFetchArgumentsTests writes the raw key and
then asserts the accessor agrees:

[[NSUserDefaults standardUserDefaults] setInteger:setting forKey:@"PBPruneOnFetch"];
XCTAssertEqual([PBGitDefaults pruneOnFetch], setting);

I had carried over the save and restore but not that cross-check.
Amended to add it:

[[NSUserDefaults standardUserDefaults] setBool:NO forKey:@"PBShowStageView"];
XCTAssertFalse([PBGitDefaults showStageView], @"the preference this suite turns off is read under some other name now, so it is no longer being turned off");

I kept the string rather than exposing a constant from PBGitDefaults,
because that turns a test-only fix into a change to production headers
for no behaviour. The cross-check buys the same protection without it.

Checked that the guard is not vacuous, by renaming the key in the test
to simulate the drift being guarded against:

• preference set, suite run as it stands: six of six pass
• key renamed: fails with
  (([PBGitDefaults showStageView]) is false) failed - the preference this suite turns off is read under some other name now

So a rename now fails at once, in every environment, rather than
dropping the defence quietly and coming back later as the same
confusing CI failure.

@hannesa2 a note on ordering, since this one blocks the others.

#693, #697 and #700 are all currently red on the same two
PBGitWindowWiringTests cases that this fixes. #686, #692 and #701 are
green, but only because their last runs built their baseline from
722a9022, which predates #695. Rebasing any of them onto current
master turns them red too, so all the open pull requests are
potentially affected and this is worth taking first.

A suggested order, if it helps:

1. This one.
2. Rebase Stop the screenshot tests spending a minute changing view #700 on top of it. It fails these exact two cases today, so it
   goes red to green and shows the fix working on the runner. That is
   worth doing on its own before the rest, because this pull request's
   own CI cannot demonstrate it: the change touches only GitXTests,
   which is not one of the paths the workflow watches for UI relevance,
   so the baseline app never launches and the preference is never set.
   The evidence is the local red to green in the description.
3. Then Support choosing the date-format for the history pane's commit date #693, Cosmetic: Fix the checked-out branch badge being dim and never updating #697 and Fix the history list still claiming to update after it was cleaned up #686 in any order.

#700 is not an alternative to this one. It makes the screenshot tests
reach their views by key equivalent rather than walking the menu bar,
which is a real improvement, but testStagingTabScreenshot still leaves
the application on the commit view, so the preference is still set
afterwards. Both are worth having.

Background is in #703. #643 is a different failure in the same test
class, still open, and #686 covers it.

( @hannesa2 , pls merge this one first - I will rebase all the other pending PRs on top of it afterwards )

Thanks, then let's start with this one

Thanks, then let's start with this one

Great!
next in line, by order of simplicity/importance:

1. Fix the history list still claiming to update after it was cleaned up #686 - Solve another CI-tests instability and performance
2. Stop the screenshot tests spending a minute changing view #700 - Improve UI-CI-tests performance
3. Cosmetic: Fix the checked-out branch badge being dim and never updating #697 - Cosmetic: Fix the colors of the V-mark of the "currently selected" branch
4. Support choosing the date-format for the history pane's commit date #693 - UI Enhancement: Support multiple date-formats in the commit-history pane