[minor] [examples] Avoid packaging duplicate classes.

[vanzin]

Add exclusions and explicit dependencies so that the examples
assembly does not duplicate classes already packaged in the main
assembly.

Also avoid relocating the commons-math3 package since it's already
a dependency of spark-core, and thus is already available in the
main assembly.

[vanzin]

Locally, with a few *-provided profiles, the examples assembly shrunk from > 80MB to about 16MB.

[SparkQA]

Test build #29763 has started for PR 5379 at commit 12c258e.

[srowen]

Individually those changes seem believable, like, marking Scala as provided and not including the Spark-shaded classes. Most affect the Cassandra dependency and the logic there is that they are definitely provided by Spark?

[SparkQA]

Test build #29763 has finished for PR 5379 at commit 12c258e.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.
• This patch does not change any dependencies.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/29763/
Test PASSed.

[vanzin]

Yes, I checked all the exclusions I'm adding; they're either direct dependencies of Spark (and thus can be provided), or are transitive (e.g. Hadoop or some other Spark dependency pulls them in), so the assembly (or the cluster where the assembly is being run) is expected to provide them.

[srowen]

This LGTM as that is a huge decrease in the size of the examples JAR. I'll leave it open for comments one more day.