Raw strings (marked by the `r` prefix) are normally used for
regular expression patterns to allow that module to expand
any backslash-escape sequences without interference from
the Python interpreter's string expansion, and thus the
requirement for extra quoting.

This practice is further encouraged by python 3.5 and later
raising a `DepreciationWarning` when encountering escape sequences
it can't expand, as is common is re patterns.

However, in python 2.7, the `re` module doesn't expand \u and \U
escapes for unicode codepoints, so patterns with non-ascii
characters can only be used in raw strings when using python 3.3
or later, or the external `regex` module, and neither `u` nor `r`
quoted strings will work for patterns in both languages.

However, Python will concatenate whitespace-separated string
literals even if they use different quoting. Take advantage
of this to split the lexer patterns into raw- and unicode-quoted
segments as appropriate.

Also remove some unnecessary escape sequences.

The pep8 name is deprecated, and the tool continues under
the new name.

Now that we're using a newer pytest, we can invoke it as such
even in the jython build, instead of using the older `py.test`.

The parser's  token pattern strings escaped almost all punctuation
characters, but the Python `re` module has a fairly limited set
of special characters which must be backslash-escaped.

In particular, characters like @ ! : and space do not need escapes,
and even more are fine inside a [] environment describing a set
of characters.

Audit the expression strings and remove most unnecessary escapes.
I left some sequences like `[^\^\[]` which aren't ambiguous without
escapes for clarity of reading.

The default osx image on travis doesn't have an up-to-date
set of homebrew packages, so the `brew upgrade` step must
compile many packages from source and takes tens of minutes
to complete. This makes for very slow test feedback and
occasionally failures when the job times out.

Travis itself doesn't provide python3 packages through
the `language` and `python` yaml keys like it does on
Linux. Setting these keys fails trying to download the
requested tarball.

Give up therefore, and just use whatever python3 was installed
on the image by default. Currently that is 3.6.5 from homebrew.
This significantly speeds up test feedback.

The #note environment can be any free text up to a double newline.
Subsequent lines don't need to be intented like with #tr.

It didn't seem like we had text coverage for this feature.

The #note environment can span multiple lines, but is supposed
to be terminated by a new line label. There was a test for
termination by an @object, but not for this.

Check all the primed variants for line numbers to prevent regression
of an issue I introduced addressing issue oracc#78.

Sanitizing the regular expression pattern strings, I accidentally
removed the optional mark ('?') from the collection of prime
characters used to conclude a #note when it doesn't end in a
double newline.

Restore this, and clarify the related comment.

Python 2.7 can't handle the unicode prime values in normal `str`
objects, so construct the generated line label as a unicode object
instead.

I used this to guide constructing the new unit tests.
Thanks to ageorgou for catching that I'd left it in
the committed version.

Rename a varable to avoid confusion with a built-in.
Thanks to ageorgou for the suggestion.