D3870: rebase: add --confirm option

[Yuya Nishihara]

>   Ah, this patch is based on https://phab.mercurial-scm.org/D3830 which is
>   not reviewed. Do I need to rebase this to hg-commited current tip?

Can you rewrite this to not depend on D3830 if that's easy? It appears
nobody reviewed D3830 yet, and I can't trivially say it's good to go.

>   > If KeyboardInterrupt is raised here for example, abort() wouldn't be called.
>   
>   okay, can I update this patch or send a follow-up to handle KeyboardInterrupt?

Can you fix that in this patch? KeyboardInterrupt is just an example. The
abort in the finally clause can't be gated by the confirm flag.