Fix checkout errors blaming the working directory (issue #629) - #630

git says why it refused, and GitX said something else.

First slice of #629. Checking out a branch that is checked out in another
worktree fails with

fatal: 'feature' is already used by worktree at '/path/to/wt'

and the sheet answered "Perhaps your working directory is not clean?", which
sends you looking for uncommitted changes that are not there. git's own text
did reach the sheet, but further down under "Output:", below a headline that
contradicted it.

The same canned guess stood in for every failure of checkoutRefish:,
checkoutFiles:fromRefish: and cherryPickRefish:, so all three now lead with
what git said. Worktrees are only the case that made it obvious.

• Report the failed task's own output as the failure reason, keeping the guess
  for a failure that produced no output (a signal or a timeout)
• Cover the worktree refusal, and both fallbacks, with tests that need no git
  on the machine

Nothing here teaches GitX about worktrees. Marking those branches and offering
to open them is the rest of #629.

Test plan

• make unit-test: 105 tests, 0 failures, 4 of them new (in
  PBGitRepositoryCheckoutErrorTests)
• Confirmed red first: with the helper returning the fallback, as the code did
  before, testGitsOwnRefusalIsReportedInsteadOfTheGuess and
  testGitsOutputIsTrimmed fail
• The tests build the task error directly rather than launching git, so they
  do not depend on [PBGitBinary path] resolving on the runner, which Keep the unit tests from depending on the machine they run on #620
  had to clean up elsewhere
• Not checked by hand in a running app

One known wart

The sheet now prints git's message twice: once as the headline, and again in
the "Output:" block that PBGitXMessageSheet appends for any task error.
Suppressing the second copy means touching the sheet for every error it shows,
which I kept out of this PR. Happy to follow up if you would rather have it.