Add metrics to KafkaClient.

[ucarion]

I am trying to collect data about how long I'm waiting on Kafka; I want to answer questions like "how much latency is there for getting messages out of Kafka?". To this end, I've put together this PR, which adds an option to enable metrics collection in KafkaClient.

This uses scales, and would let me write stuff like this:

from kafka import KafkaConsumer

consumer = KafkaConsumer('foobar',
                         bootstrap_servers=['localhost:9092'],
                         group_id='kafka_test',
                         enable_metrics=True)

# interesting logic would go here ...
for _ in xrange(10):
    for message in consumer.fetch_messages():
        process(message)

# outputs the 95th percentile of how long it takes to perform send_fetch_request
print consumer.metrics.fetch_request_timer['95percentile']

I've only added support for metrics in KafkaConsumer, but I can add other consumers if you agree with this design.

[patricklucas]

Check out functools.wraps.

[ecanzonieri]

It may also be useful to get errors count as part of these metrics

[dpkp]

the upstream client configuration here is metric.reporters. I try to stay consistent, so perhaps metric_reporters and support a list of reporter interfaces?

[dpkp]

Metrics support would be great -- if you have time, take a look at my comments re: consistency w/ upstream client approach. Otherwise I'll probably put this off for a bit and clean up w/ a larger 0.9 client refactor.