bug#57400: 29.0.50; Support sending patches from VC directly

[Juri Linkov <juri@linkov.net>]

> Again, regarding `crm-separator' I don't know if this is a value that
> people change or not (perhaps it should be changed to a defconst).

At least, it's not defcustom, so it's not intended for user customization.

> I can also imagine that the initial input outside of a log buffer ought
> to just be the last commit.

Depends on whether this is one of the most popular workflows.

>> Additional question: shouldn't the behavior of
>> vc-prepare-patches-separately=nil be equivalent to
>> vc-prepare-patches-separately=t when only one revision is given?  Both
>> cases create just one mail.  So when vc-prepare-patches-separately is
>> nil, could it extract the subject from the single revision?  Or maybe
>> vc-prepare-patches-separately should support a new customization value
>> for this?
>
> Yes, but the formatting is different.  In the one case you attach a
> patch to a regular message, while in the other case the message is
> prepared in such a way that (at least when using Git) the entire message
> is a valid patch, where we can use "git am".

Hmm, I tried both variants, and still not sure which is better
for the 1-patch case.  However, what I definitely suggest to do is
to get rid of recursive-edit, that also Robert noticed on emacs-devel.
recursive-edit is too fragile for modal editing, because such
commands as keyboard-escape-quit can easily break it.  Without
recursive-edit you can just create all mail buffers at once.
Then after sending one mail, its buffer gets buried, and the
next mail buffer will be shown instead, etc.