[SPARK-16602][SQL] `Nvl` function should support numeric-string cases

## What changes were proposed in this pull request?

`Nvl` function should support numeric-straing cases like Hive/Spark1.6. Currently, `Nvl` finds the tightest common types among numeric types. This PR extends that to consider `String` type, too.

```scala
- TypeCoercion.findTightestCommonTypeOfTwo(left.dataType, right.dataType).map { dtype =>
+ TypeCoercion.findTightestCommonTypeToString(left.dataType, right.dataType).map { dtype =>
```

**Before**
```scala
scala> sql("select nvl('0', 1)").collect()
org.apache.spark.sql.AnalysisException: cannot resolve `nvl("0", 1)` due to data type mismatch:
input to function coalesce should all be the same type, but it's [string, int]; line 1 pos 7
```

**After**
```scala
scala> sql("select nvl('0', 1)").collect()
res0: Array[org.apache.spark.sql.Row] = Array([0])
```

## How was this patch tested?

Pass the Jenkins tests.

Author: Dongjoon Hyun <dongjoon@apache.org>

Closes #14251 from dongjoon-hyun/SPARK-16602.