Add a list of provided library dependencies

[omervk]

A very long classpath including user jars is loaded when a Spark node starts. Instead of users having to figure out which jars are in memory whenever they change a Databricks cluster type / version, there should be a key of ProvidedDependencies for each distinct version.