Add some basic request metrics

[ewencp]

Add some basic metrics that work for any rest-utils application. The two commits add metrics at different levels. The first adds some basic TCP connection-level stats by hooking into Jetty, and the second adds HTTP level stats by hooking into Jersey.

It's probably easiest to see what there is by looking at the JMX stats for the example application. A few important questions for review:

1. What else should we include by default, i.e. are any key metrics still missing? More metrics to add on existing sensors? Latency 95th/99th/99.9th percentiles are the obvious ones I can think of, but it's unclear to me how good the current implementation of percentiles is.
2. Metrics names/organization. The metrics library has some support for hierarchical Sensors with aggregation, but the ordering of names felt backwards. It led to names like request-rate.TopicResource.produce so that all requests could be aggregated into request-rate, but that seemed wrong when I also wanted multiple metrics grouped under TopicResource.produce. The current patch just has extra sensors that are always used for the global stats.
3. Names are always autogenerated via class names and method names. Is this what we want, or should we allow/encourage/force better naming via attributes?

[ewencp]

@nehanarkhede Thoughts on this? Ideally we can get a good set to cover most of what both kafka-rest and schema-registry need.

[ewencp]

Blocked on confluentinc/common#11

[ewencp]

@nehanarkhede Ready for you to look at this again. confluentinc/kafka-rest#23 shows how it gets used. There are a couple of changes required because I had to rework some rest-utils classes (we'll also have to make them in schema-registry), but the major change to users of rest-utils is just annotating the resource methods with better names.

[ewencp]

Merging based on review of confluentinc/kafka-rest#23.