Fix prometheus passthrough for existing value types

[danielnelson]

This is an alternative pull request to #3337, I added the setting of metric types in the prometheus input, and use that in the prometheus output to adjust the output format. I think this will have less potential for side effects.

I also included @jdoupe suggestion to use the promhttp package, as well as adding his configurable disabling of the built in collectors. It would be nicer to disable them by default, since outputs shouldn't really create metrics, but for now they default on.

I added the ErrorHandling: promhttp.ContinueOnError option which seems to prevents conflict if the help message is different. This will reduce the visibility of errors, but it seems it may be necessary. In the long term I would like to add metadata support to our metrics struct, so that help info can be passed through this way, but this requires more consideration.

Unlike #3337, I did not add histogram or summary support, these types need to be added as first class telegraf.ValueType types and then they can be wired up like counters/gauges.

closes #2950

Required for all PRs:

• Signed CLA.
• Associated README.md updated.
• Has appropriate unit tests.

[jdoupe]

Here's an ugly patch (not that ugly, but probably needs some cleanup) that is built on top of #3351 and adds history and summary types. https://github.com/influxdata/telegraf/compare/master...jdoupe:add-history-and-summary-types?expand=1

[danielnelson]

I'll merge this and then can you rebase and open a pull request?