auth_token middleware fails for concurrent requests

We were trying to run some performance tests on [melange](https://code.launchpad.net/~raxnetworking/nova/melange) using [keystone](https://github.com/openstack/keystone) auth_token middleware. The auth_token middleware fails with "500 Internal Server Error" for concurrent requests. I have pasted the server log below.

```log
2011-08-10 14:36:56 DEBUG [eventlet.wsgi.server] Traceback (most recent call last):
  File "/home/rackspace/openstack/nova_melange/.nova-venv/lib/python2.6/site-packages/eventlet/wsgi.py", line 345, in handle_one_response
    'Content-Length' not in [h for h, _v in headers_set[1]]:
IndexError: list index out of range
```

The log just gives an indication that heders_set array is empty. Digging deep in to the error we found that this was due to the auth_token middleware.

This happens because *start_response* callback is stored as a instance level attribute in auth_token and being used in serveral private methods. When concurrent request are made, same instance of start_response callback is being reused across the multiple request.

The issue can be fixed by passing start_response as arguments to these private_methods such as _forward_request etc.

The same applies to request *env* and *proxy_headers* variables, which are being used as instance attributes should be passed as method arguments.

If you need any help in reproducing issue, please let me know. I have pasted the command that was used for benchmarking.

```
autobench --single_host --host1=10.10.253.25 --port1=9898 --uri1=/v0.1/ipam/ip_blocks --low_rate=50 --rate_step=10 --high_rate=60 --num_call=1 --num_conn=100 --file=results.tsv --timeout=180
```
Response:

```
Connection rate: 0.9 conn/s (1080.7 ms/conn, <=100 concurrent connections)
.
.
Reply status: 1xx=0 2xx=1 3xx=0 4xx=0 5xx=99
```