This should fix the tests failing on PyPy. Eventually we'll need a more
robust solution for this.

Explicitly handle unclosed <script> and <style> tags which previously
would result in O(n^2) work to lex as Error tokens per character up to
the end of the line or end of file (whichever comes first).

Now we try lexing the rest of the line as Javascript/CSS  if there's no
closing script/style tag. We recover on the next line in the root state
if there is a newline, otherwise just keep parsing as Javascript/CSS.

This is similar to how the error handling in lexer.py works except we
get Javascript or CSS tokens instead of Error tokens. And we get to the
end of the line much faster since we don't apply an O(n) regex for every
character in the line.

I added a new test suite for html lexer (there wasn't one except for
coverage in test_examplefiles.py) including a trivial happy-path case
and several cases around <script> and <style> fragments, including
regression coverage that fails on the old logic.

* testing turtle prefix names where reference starts with number

* remove case insensitive flag from Turtle lexer

* use same end-of-string regex as in SPARQL and ShExC

* make example.ttl valid turtle

* JavaLexer: Demonstrate a catastrophic backtracking bug

* JavaLexer: Fix a catastrophic backtracking bug

Closes #1586

Previously, the tag was cut off.

Previously, something like:
<%class>text</%class> would not get matched correctly. This was due to
the capturing group capturing the wrong part of the tag -- instead of
class, it would capture the part after class before >. With this commit,
the capturing group correctly matches the start/end tag. This commit
also adds a unit test to verify this.