commons-lang (2.x) classes used instead of commons-lang3 (3.x); commons-io used without dependency

Spark depends only on Commons Lang3 (3.x). However there are several accidental usages of Commons Lang (2.x) in the codebase. As we saw a few days ago the version of Lang 2.x that accidentally comes in via Hadoop can change with Hadoop version and so the accidental usage is more than a purely theoretical problem. It's easy to change the usages to 3.x counterparts.

Also, there are just a few uses of Commons IO in the code which can be replaced with uses of Guava, removing another used but undeclared dependency.