initCap behavior mismatches Spark semantics

[yew1eb]

Describe the bug
The current initcap implementation uses DataFusion's initcap, which does not match Spark's semantics.
Spark's initcap function semantics (per its description):

Returns `str` with the first letter of each word in uppercase. 
All other letters are in lowercase. Words are delimited by white space.

DataFusion's initcap function semantics (per its description):

Capitalizes the first character in each word in the input string. 
Words are delimited by non-alphanumeric characters.  

#1549

To Reproduce
Run the follwing SQLs:

CREATE TABLE  tbl(id INT, txt STRING) USING parquet;
INSERT INTO tbl VALUES
(1, 'Hello_world'),
(2, 'robert rose-smith'),
(3, 'foo.bar/baz')
 ;

select id, initcap(txt) from tbl;  

Expected behavior

1,Hello_world
2,Robert Rose-smith
3,Foo.bar/baz

Actual behavior

1,Hello_World
2,Robert Rose-Smith
3,Foo.Bar/Baz

Screenshots

Additional context

[xumingming]

@yew1eb Good catch! Just curious how do you find all these mismatch issues? And seems Auron needs to do more on the query correctness things.