Make it easy to find out what version of the program is installed.

The French attribution from before capitalised the month name because I
have an EN locale by default. The month shouldn't be capitalised in
French. Set the `$attribution_locale` to match the language of the
attribution template string.

Hard-coding `en_US` on the English one as that's the locale I'm used to.

Now that we've handled 'multipart/relative' and HTML-only
'text/plain'-less emails, we can remove these.

Now that we're using it in two places, move it to a function. Also
remove the `unwrap` call.

We're already doing this for single-part emails. This change makes it
work additionally for multipart emails.

Now that I know the `Content-Type` header is already parsed in the
original email parse call, use that instead of looping over the headers
trying to find the content type.

When an HTML body is fed to 'whatlang', it recognises it as English.
This is likely due to the English HTML syntax. Remove all HTML tags with
a simple regex substitution to get the language recognition working more
properly.

This doesn't remove CSS, which could also confuse the language
recogniser. In a limited test, it seemed to work without having to
remove any CSS, so not bothering with that.

Still need to get this working for multipart emails.

Figuring out the different content types and subparts for different
kinds of emails.

Parse the text email body out of `multipart/alternative` parts.