[SPARK-18772][SQL] Avoid unnecessary conversion try for special floats in JSON

## What changes were proposed in this pull request?

This PR is based on  https://github.com/apache/spark/pull/16199 and extracts the valid change from https://github.com/apache/spark/pull/9759 to resolve SPARK-18772

This avoids additional conversion try with `toFloat` and `toDouble`.

For avoiding additional conversions, please refer the codes below:

**Before**

```scala
scala> import org.apache.spark.sql.types._
import org.apache.spark.sql.types._

scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show()
17/05/12 11:30:41 ERROR Executor: Exception in task 0.0 in stage 2.0 (TID 2)
java.lang.NumberFormatException: For input string: "nan"
...
```

**After**

```scala
scala> import org.apache.spark.sql.types._
import org.apache.spark.sql.types._

scala> spark.read.schema(StructType(Seq(StructField("a", DoubleType)))).option("mode", "FAILFAST").json(Seq("""{"a": "nan"}""").toDS).show()
17/05/12 11:44:30 ERROR Executor: Exception in task 0.0 in stage 0.0 (TID 0)
java.lang.RuntimeException: Cannot parse nan as DoubleType.
...
```

## How was this patch tested?

Unit tests added in `JsonSuite`.

Closes #16199

Author: hyukjinkwon <gurwls223@gmail.com>
Author: Nathan Howell <nhowell@godaddy.com>

Closes #17956 from HyukjinKwon/SPARK-18772.