[SPARK-4953][Doc] Fix the description of building Spark with YARN

[sarutak]

At the section "Specifying the Hadoop Version" In building-spark.md, there is description about building with YARN with Hadoop 0.23.
Spark 1.3.0 will not support Hadoop 0.23 so we should fix the description.

[SparkQA]

Test build #24769 has finished for PR 3787 at commit 264e4e0.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

This is not only applicable to CDH 5+, so I'd revert that addition. What was removed with yarn-alpha was not really "Hadoop 0.23" support, although it kind of lines up with that. Why not remove this whole qualifying "For Apache Hadoop ..." phrase altogether? Also, do you mean Spark 1.2? what are you referring to in 1.3 otherwise?

[SparkQA]

Test build #24771 has finished for PR 3787 at commit 9ab0c24.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • case class ApplicationFinished(id: String)

[srowen]

This has a typo in "CDEH" and are also not runnable. I don't see much value in elaborating this example 3 more times.

(As a related aside, I would like to see less, not more, vendor stuff in Spark anyway. Adding just this text unduly favors Cloudera a tiny bit; the alternative is to write a bunch of other vendor combos here, which is going to turn into at least a maintenance headache. I already disagree with maintaining vendor versioning info in the project POM.)

[SparkQA]

Test build #24783 has finished for PR 3787 at commit ee9c355.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[pwendell]

This looks good - thanks @sarutak and @srowen!