Implement ARRAY_NORMALIZE function

[yuanzhanhku]

fixes #16134

Normalizes array x by dividing each element by the p-norm of the array.
It is equivalent to
TRANSFORM(array, v -> v / REDUCE(array, 0, (a, v) -> a + POW(ABS(v), p), a -> POW(a, 1 / p)).
But the reduce part is only executed once to improve performance.
Returns null if the array is null or there are null array elements.

Test plan

• Added unit tests in src/test/java/com/facebook/presto/operator/scalar/TestArrayNormalizeFunction.java.
• Also started a presto server locally and tested the change manually.

== RELEASE NOTES ==

General Changes
* Added array_normalize function

[linux-foundation-easycla]

The committers are authorized under a signed CLA.

• ✅ Zhan Yuan (f50c6ed85f489bcea4f3f0f3e7e529e9c00bb813)

[highker]

nits only

[highker]

one more nit

[kaikalur]

I wonder if this should work like an "Aggregation" so ignore nulls?

[yuanzhanhku]

Good question. Before I prepared this commit, I discussed with erikbrinkman who requested this function in issue #16134. There was no specific requirement for how to handle null element.

Technically, this is not an Aggregation function as it returns an array instead of a scalar value. Thus there is no absolute right and wrong answers for this custom function. We just need to make sure the function comment states the behavior correctly. If you think ignoring nulls is better, I am happy to change this behavior.

[kaikalur]

This particular thing of every operation adding their own classes for typed accessors is becoming too much code bloat. See if you can make it some kind of utility class at the top level. For example, I had to do something similar in MapUnionSum:

presto/presto-main/src/main/java/com/facebook/presto/operator/aggregation/state/MapUnionSumStateFactory.java

Line 213 in 9095346

type.writeLong(blockBuilder, BigintOperators.add(type.getLong(block1, position1), type.getLong(block2, position2)));

[yuanzhanhku]

Totally agree. I took a closer look at the Type interface. I think it would be possible to move the type specific accessors into the subclasses of Types.

[yuanzhanhku]

Prepared #16209 to address this. Please see if the change makes sense.