[SPARK-10540] [SQL] Ignore HadoopFsRelationTest's "test all data types" if it is too flaky

[yhuai]

If hadoopFsRelationSuites's "test all data types" is too flaky we can disable it for now.

https://issues.apache.org/jira/browse/SPARK-10540

[SparkQA]

Test build #42284 has finished for PR 8705 at commit 7c4791f.

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

[liancheng]

Found two possible sources of flakiness:

1. For double and float, NaN == NaN is always false

2. checkAnswer doesn't handle Map[K, V] properly. For example:

   scala> Map(1 -> 2, 2 -> 1).toString
   res0: String = Map(1 -> 2, 2 -> 1)
   
   scala> Map(2 -> 1, 1 -> 2).toString
   res1: String = Map(2 -> 1, 1 -> 2)
   

   We can't rely on toString to compare Map instances.

Need to update checkAnswer to special case NaN and Map[K, V].

[liancheng]

Filed https://issues.apache.org/jira/browse/SPARK-10591 to track this issue.

[yhuai]

For the issue of NaN, looks like there is something wrong because we want to return true when compare two NaN.