Delaying 2.0 final

As we seem to have a few regressions we may want to fix, I will not be cutting the 2.0 final today (https://tinyurl.com/gitCal).

I queued the following near the bottom of 'pu' (these are also merged to 'next' to keep pu^{/match.next} in sync with next), and plan to cut 2.0.0-rc4 early next week.

 * git-gui has a show-stopper bug that wants to see if it is used
   with a version of git that "vsatisfies" 1.7.0; because 2.x is
   considered significantly different, it incorrectly decides to use
   the codepath for ancient versions that compares older than 1.7.0
   when working inside a submodule working tree.  A fix by Jens
   Lehmann is queued as an emergency patch to use vcompare instead.

 * There was a vague admission of having broken git-remote-hg/bzr by
   its author that hinted one topic branch is the culprit, without
   much detail.  As a precautionary measure, the merge of the topic
   has been reverted.

 * git-remote-hg/bzr are maintained outside our tree, our copy will
   become stale and will eventually go away.  A warning message to
   its standard error output has been added to each of them to tell
   the users where to find the latest of these tools.

 * git-request-pull has a minor regression and lost a useful DWIM
   that was designed to be friendly to users of Git older than
   1.7.10.  "git request-pull $base $URL for-linus" used to say
   "Please pull tags/for-linus" if that is what the user meant, but
   the version in 2.0.0-rc3 requires the command line to be "git
   request-pull $base $URL tags/for-linus".  It is simple enough to
   special case this use case, and a patch is included to fix it
   with a band-aid.

If you fetch tonight's 'pu' branch, you will find all of the above at f608b83. Hopefully 2.0.0-rc4 and 2.0.0-final will be the same tree as that commit, with updated release notes.