use histogram for http_request_duration_seconds

[tmc]

closes #27

[coveralls]

Coverage remained the same at 98.819% when pulling 8785389 on tmc:tmc.histogram into 40091e1 on prometheus:master.

[grobie]

So the tests would need to get updated, but I was thinking about changing the default metric name according to @brian-brazil's http://www.robustperception.io/on-the-naming-of-things/. Given this will be a breaking change anyway, we could do both things at once.

How awkward is the namespacing? Would it make more sense to move the rack middlewares to something like Prometheus::Rack::Exporter or a more generic Prometheus::Middleware::Exporter?

[grobie]

Replaced by #32