Remove use of spark.worker.instances

spark.worker.instances was added as part of this commit: https://github.com/apache/spark/commit/1617816090e7b20124a512a43860a21232ebf511

My understanding is that SPARK_WORKER_INSTANCES is supported for backwards compatibility,
but spark.worker.instances is never used (SparkSubmit.scala sets spark.executor.instances) so should
not have been added.

@sryza @pwendell @tgravescs LMK if I'm understanding this correctly

Author: Kay Ousterhout <kayousterhout@gmail.com>

Closes #1214 from kayousterhout/yarn_config and squashes the following commits:

3d7c491 [Kay Ousterhout] Remove use of spark.worker.instances

yarn/common/src/main/scala/org/apache/spark/scheduler/cluster/YarnClientSchedulerBackend.scala

@@ -63,7 +63,7 @@ private[spark] class YarnClientSchedulerBackend(
     // variables.
     List(("--driver-memory", "SPARK_MASTER_MEMORY", "spark.master.memory"),
       ("--driver-memory", "SPARK_DRIVER_MEMORY", "spark.driver.memory"),
-      ("--num-executors", "SPARK_WORKER_INSTANCES", "spark.worker.instances"),
+      ("--num-executors", "SPARK_WORKER_INSTANCES", "spark.executor.instances"),
       ("--num-executors", "SPARK_EXECUTOR_INSTANCES", "spark.executor.instances"),
       ("--executor-memory", "SPARK_WORKER_MEMORY", "spark.executor.memory"),
       ("--executor-memory", "SPARK_EXECUTOR_MEMORY", "spark.executor.memory"),