This commit simplifies async/await tokenization in tokenizer.c,
tokenize.py & lib2to3/tokenize.py.  Previous solution was to keep
a stack of async-def & def blocks, whereas the new approach is just
to remember position of the outermost async-def block.

This change won't bring any parsing performance improvements, but
it makes the code much easier to read and validate.

* Tests that index.html is served, rather than an automatic directory listing
* Tests that there is no extra data sent after the response

Patch by Martin Panter.

This commit fixes how one-line async-defs and defs are tracked
by tokenizer.  It allows to correctly parse invalid code such
as:

>>> async def f():
...     def g(): pass
...     async = 10

and valid code such as:

>>> async def f():
...     async def g(): pass
...     await z

As a consequence, is is now possible to have one-line
'async def foo(): await ..' functions:

>>> async def foo(): return await bar()

Patch by Jacek Kołodziej.

Patch by Jacek Kołodziej.

Patch by Kai Groner.

According to the documentation, they must be decorated as classmethods.

Original patch by Saimadhav Heblikar.

for single-byte argument on Linux.

Patch written by Martin Panter. I replace tearDown() with addCleanup().

explicitly forbidden as well as numeric group references.

Patch by Brian Cain.