Add support for multiple monitoring endpoints per scraped pod

[razvan]

The problem

Currently only one metrics path per scraped pod is supported. Usually this is called /metrics and is configurable for each scraping job. Some systems, like Apache Spark use two different paths to separate cluster from application metrics. See metric template

Proposed solution

There are at least two possible solutions. The easy, fast and ugly one and the more elegant but complicated to reach one.

Solution 1: Configure an additional scraping jobs for each metrics not available through the /metrics endpoint. For Spark masters, this means an additional job for application metrics. The disadvantage is that these jobs are always present, even in clusters where their targets are not even deployed.

Solution 2: Configure scraping jobs dynamically based on what is actually deployed and running in the K8S cluster. This is what the official Prometheus operator is doing.

[lfrancke]

We decided to use the Prometheus Operator instead