implement 'StringConcat' operator to support sql like "select 'aa' || 'b' "

[WinkerDu]

Which issue does this PR close?

Closes #2141 .

Rationale for this change

df now has not implemented StringConcat operator, like

❯ select 'aa' || 'b';
NotImplemented("Unsupported SQL binary operator StringConcat")

But Postgres SQL will come out results like

select 'aa' || 'b';
 ?column? 
----------
 aab
(1 row)

What changes are included in this PR?

Reuse df build-in concat string expression to do it.

Are there any user-facing changes?

No.

[WinkerDu]

cc @Dandandan @alamb @xudong963 @yjshen
Please have a review, thank you.

[jackwener]

Great job! ❤

[WinkerDu]

@jackwener thank you for your comment.
I'm not sure what PG behaves like is reasonable, PG build-in expression concat can take NULL like

select concat('aa', NULL, 'b');
 concat 
--------
 aab
(1 row)

I prefer unify the outputs of || and concat, what other contributors think of?

[WinkerDu]

@jackwener thank you for your comment. I'm not sure what PG behaves like is reasonable, PG build-in expression concat can take NULL like

select concat('aa', NULL, 'b');
 concat 
--------
 aab
(1 row)

I prefer unify the outputs of || and concat, what other contributors think of?

@jackwener @renato2099
I've check the behaviors of PG of SparkSQL, they all come out results with NULL

• Spark 3.2

spark-sql> select 'a' || NULL || 'b';
NULL

• PG

# select 'a' || NULL || 'b';
 ?column? 
----------
 
(1 row)

Seems here is a convention, I'll comply with it in later commit, thank you both.

[alamb]

Thanks @WinkerDu -- this is a nice addition

[wesm]

@alamb do you know why this PR generated e-mails to the dev@ mailing list?

[alamb]

@wesm I am sorry but do not know why there were some emails from this PR to dev e. g. https://lists.apache.org/thread/tqv2x57xmzzrppl3zgox5rt2rhfyg8jz

I vaguely remember getting an email from apache infra about "remailer" but I didn't read the details and now I can't seem to find it :(

There appears to be at least one C++ github conversation that got forwarded too: https://lists.apache.org/thread/vr3yjrcrb01b8zh64dclfl4d2gcxfw0j

[wesm]

It seems to have been a temporary INFRA hiccup (Iceberg had some issues, too https://lists.apache.org/thread/8f7xdfttk4xhfo7o66lwkpc4xzvx2vjz). If it starts happening again, we should open an INFRA Jira ticket.

[WinkerDu]

@alamb @jackwener @renato2099 PTAL, thank you.

1. support concat string and values that can be cast to string, add some UT, like select 'aa' || 42.
2. output NULL when encounter NULL in concat pipe.
3. keep StringConcat operator as BinaryExpr to make column physical name correct.

[WinkerDu]

@alamb PTAL, thank you.

[alamb]

LGTM -- thanks @WinkerDu

[alamb]

As a future optimization it might be worth using the take kernel https://docs.rs/arrow/11.1.0/arrow/compute/kernels/take/fn.take.html which is fairly well optimized / tries to avoid copying.

This is good for now though