It is better readable.

Make sure signed entries are stripped also if dash-escaped.

Create a single place where lines of wordlists can be refined,
regardless of their mode (crypt-signed or not).

We ignore empty lines also in signed wordlists. To check it, we need
an empty line in the sample.

Many wordlists have format "NNNNNN<tab>some-term". We try to
detect lines of this format and will parse them to get the term
(removing numbers and whitespaces).

and do not strip() if done already.

We need this regular expression to strip entries in wordlists like
the original 7776-terms wordlist from diceware.com.

As described in RFC 4880, dashes in wordlists might be escaped.
We have to take care of this when extracting wordlists.

In signed files, words with a starting dash are treated special
(they are "dash-escaped" as RFC 4880 says).

This sample wordlist contains only 2 entries but is
cryptographically signed. We use it to make sure we can also
handle signed wordlists.

The changed test could fail if the wordlist dir was relatively short.

We expect more wordlist-specific changes, regarding signatures etc.

This one is best suited for real dice passphrases.

and bring coverage back to 100%.

The temporary wordlists dir must be a string.

With this fixture we can test different wordlist file names without
touching the real wordslists dir. This is important as we cannot say
what files a user stores or removes from that originial directory.

The README is evaluated by standard docutils when released on
pypi.python.org. These do not understand most text roles (like
``:class:``) recognized by `Sphinx`. As a result the README will be
displayed as raw text on PyPI which looks ugly.

There is no need for us to enforce lowercase wordlist names. It
could even confuse people.