[SPARK-19919][SQL] Defer throwing the exception for empty paths in CSV datasource into `DataSource`

## What changes were proposed in this pull request?

This PR proposes to defer throwing the exception within `DataSource`.

Currently, if other datasources fail to infer the schema, it returns `None` and then this is being validated in `DataSource` as below:

```
scala> spark.read.json("emptydir")
org.apache.spark.sql.AnalysisException: Unable to infer schema for JSON. It must be specified manually.;
```

```
scala> spark.read.orc("emptydir")
org.apache.spark.sql.AnalysisException: Unable to infer schema for ORC. It must be specified manually.;
```

```
scala> spark.read.parquet("emptydir")
org.apache.spark.sql.AnalysisException: Unable to infer schema for Parquet. It must be specified manually.;
```

However, CSV it checks it within the datasource implementation and throws another exception message as below:

```
scala> spark.read.csv("emptydir")
java.lang.IllegalArgumentException: requirement failed: Cannot infer schema from an empty set of files
```

We could remove this duplicated check and validate this in one place in the same way with the same message.

## How was this patch tested?

Unit test in `CSVSuite` and manual test.

Author: hyukjinkwon <gurwls223@gmail.com>

Closes #17256 from HyukjinKwon/SPARK-19919.