[SPARK-22435][SQL] Support processing array and map type using script

[jinxing64]

What changes were proposed in this pull request?

Currently, It is not supported to use script(e.g. python) to process array type or map type, it will complain with below message:
org.apache.spark.sql.catalyst.expressions.UnsafeArrayData cannot be cast to [Ljava.lang.Object
org.apache.spark.sql.catalyst.expressions.UnsafeMapData cannot be cast to java.util.Map

This pr proposes to support it by using DelimitedJSONSerDe
This pr also fixes a bug -- when using input row format with script, no data will be produced from ScriptTransformationExec.

How was this patch tested?

Tests added.

[jinxing64]

t.getText doesn't work, we need to process the token. e.g. remove the quote

[jinxing64]

writer writes records into the input stream of script. Isn't it should be initialized with input format?

[jinxing64]

DelimitedJSONSerDe doesn't support deserialize.
Use DelimitedJSONSerDe for input SerDe and LazySimpleSerDe for output SerDe.

[jinxing64]

TODO: build json string for more types

[SparkQA]

Test build #83399 has finished for PR 19652 at commit 4a99426.

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

[SparkQA]

Test build #83446 has finished for PR 19652 at commit 0d706ff.

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

[gatorsmile]

Thanks for working on it. Will review it next week.

[jinxing64]

@gatorsmile
(Very gentle ping)
Could you please give some comments when you have time :)
Thanks you so much :)

[SparkQA]

Test build #100089 has finished for PR 19652 at commit 0d706ff.

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