[commit-patch] Use Monotone's --without-header option instead of looking for diffs that are exclusively comments.

[commit-patch] Avoid running 'interdiff' and 'patch -R' also when the working patch is just a series of comment lines.

'mtn diff' will output three comment lines when there are no changes.
patch ignores them, so should we.

[commit-patch] Use lsdiffs -s option to keep track of added, removed, and changed files in the patch so we can use them appropriately.

[commit-patch] VCS Add and Remove needs to happen after the patch has been applied to the directory.

Thanks to Richard Levitte <richard@levitte.org> for pointing this out.

Patch fails when the input has just a single \n in it which is what happens when
we do "darcs diff a" and "a" doesn't exist yet.

It's cleaner and makes it easy to look for empty files (next patch) but it
doesn't scale as well since both working and non-committed patches have to fit
in memory (which probably doesn't matter since most patches aren't massive).

[commit-patch-buffer.el] Mention commit-patch the script in the help text so the connection to the script is explicit.

[commit-patch-buffer.el] Accept a prefix parameter to commit-patch-buffer which makes it ask for the buffer and directory, like old times.

I think it's a better, more descriptive name.

[commit-patch] Add --retry option to commit-partial for when the edited patch fails apply the first time.

This avoids tyring to do diff-hunk-next in autodetect-patch-directory-root
and is more flexible all around.

There are now just 2 functions:

commit-patch-buffer - Fully automatic, takes no arguments (bound to C-c C-c).
commit-patch-buffer-manually - Takes 2 interactive arguments.

Dropped commit-patch-current-buffer which took just a directory argument as
it seems pointless now--use commit-patch-buffer-manually instead.

This takes commit-patch down to zero arguments!

[commit-patch-buffer.el] Add commit-patch-current-buffer which skips the "ask for buffer to patch" step.

So far in all my years of commit-patch usage I don't think I've ever needed
to commit a buffer that I wasn't looking at right then.

END always gets run, even when we exit via signals and we have a handler.