v2 _notmuch_message_ensure_metadata exception handling

[Jani Nikula]

On Sat, 25 Feb 2017, David Bremner <david at tethera.net> wrote:
> I tried to address most of Jani's comments on v1 [1]. The bad news is
> that the test is a bit delicate, it really does need the whole corpus.

Too bad. I hope it's not too dependent on the environment and the phase
of the Moon, etc.

> I did decide to follow the suggestion to make the _reopen function
> private for now. It would be nice if we could hide that complexity
> from library users, I guess we'll have to see.

It's easy enough to expose that later if need be.

The series LGTM.

BR,
Jani.