cope with inline PGP encrypted messages

[Daniel Kahn Gillmor <dkg@fifthhorseman.net>]

Inline PGP encrypted messages are clearly worse than PGP/MIME
structured encrypted messages.  There are no standards for how they
are formed, and they don't offer any structured metadata about how to
interpret the bytestream produced by decrypting them.

However, some other MUAs and end-user workflows may make creation of
inline PGP encrypted messages the only available option for message
encryption, and when Notmuch encounters such a message, it should make
a reasonable best-effort to render the cleartext to the user.

Due to ambiguities in interpretation of signatures on inline messages
(e.g. which parts of the message were actually signed?  what character
encoding should the bytestream be interpreted as), we continue to
ignore inline-signed messages entirely, and we do not look at the
validity of any signatures that might be found when decrypting inline
PGP encrypted messages.

We make use here of GMime's optimization function for detecting the
presence of inline PGP encrypted content, which is only found in GMime
3.0 or later.

This series is currently based n top of the "notmuch show
--decrypt=stash" series, which it needs to be able to apply cleanly.
If that series proves controversial, i could rebase this patch
manually against some earlier commit.

If you have applied this series, and you know you have some inline PGP
messages already in your message store, you can try to retroactively
reindex them with something like:

    notmuch reindex --decrypt=true BEGIN-PGP-MESSAGE and not tag:encrypted

I welcome review and feedback about this series.

  --dkg