This is definitely specific to rebase and how it fiddles with ancestry. The
patch that fixes your rebase case breaks a bunch of tests in the test suite.

Also, in your test case, this should -not- appear:

  local changed dd/a which remote deleted
 use (c)hanged version or (d)elete? c

..because remote didn't delete the file, it renamed it. The new code detects
the rename(!), but detects it as being a divergent rename (both sides
renaming it from its original name) because it's not following the "virtual
history" that rebase tries to impose on the merge.