fix(clickhouse): correctly transpile arrayMap and arrayFilter higher-order functions [CLAUDE]

[gaoflow]

Problem

ClickHouse's higher-order array functions arrayMap(lambda, arr) and
arrayFilter(lambda, arr) were silently becoming Anonymous expressions
during parsing because the ClickHouse parser lacked ARRAYMAP / ARRAYFILTER
entries in its FUNCTIONS dict. Consequently:

# Before
sqlglot.transpile("SELECT arrayMap(x -> x + 1, arr) FROM t",
                  read="clickhouse", write="duckdb")
# → "SELECT ARRAYMAP(x -> x + 1, arr) FROM t"   ← invalid DuckDB

sqlglot.transpile("SELECT arrayFilter(x -> x > 0, arr) FROM t",
                  read="clickhouse", write="duckdb")
# → "SELECT ARRAYFILTER(x -> x > 0, arr) FROM t"  ← invalid DuckDB

Additionally, the ClickHouse generator had no exp.Transform entry, so
LIST_TRANSFORM (DuckDB) and TRANSFORM (Spark/Hive) arriving at the
ClickHouse writer produced TRANSFORM(arr, lambda) instead of
arrayMap(lambda, arr).

Fix

• Parser (sqlglot/parsers/clickhouse.py): add "ARRAYMAP" and
  "ARRAYFILTER" to the ClickHouse FUNCTIONS mapping. Both builders swap
  the argument order since ClickHouse puts the lambda first, opposite to the
  canonical exp.Transform(this=arr, expression=lambda) convention.

• Generator (sqlglot/generators/clickhouse.py): add
  exp.Transform → arrayMap(lambda, arr) to the ClickHouse TRANSFORMS dict,
  matching the reversed argument order used by exp.ArrayFilter already.

After

sqlglot.transpile("SELECT arrayMap(x -> x + 1, arr) FROM t",
                  read="clickhouse", write="duckdb")
# → "SELECT LIST_TRANSFORM(arr, x -> x + 1) FROM t"  ✓

sqlglot.transpile("SELECT arrayFilter(x -> x > 0, arr) FROM t",
                  read="clickhouse", write="duckdb")
# → "SELECT LIST_FILTER(arr, x -> x > 0) FROM t"  ✓

sqlglot.transpile("SELECT LIST_TRANSFORM(arr, x -> x + 1) FROM t",
                  read="duckdb", write="clickhouse")
# → "SELECT arrayMap(x -> x + 1, arr) FROM t"  ✓

sqlglot.transpile("SELECT arrayMap(x -> x + 1, arr) FROM t",
                  read="clickhouse", write="clickhouse")
# → "SELECT arrayMap(x -> x + 1, arr) FROM t"  ✓ (round-trip)

All 1127 existing tests continue to pass.

---

This pull request was prepared with the assistance of AI, under my direction and review.

[geooo109]

Nice work, I left some comments.

My suggestion is the following:

Remove the current validate_all tests of this PR, and only use vaildate_identity tests for testing the roundtrip (input clickhouse -> output clickhouse) of the added functions for example self.validate_identity("arrayFilter(x -> x > 0, arr)").assert_is(exp.ArrayFilter). This would make the PR simple and mergable.

Then, if you want you can investigate various inputs - outputs for the transpilation, as you saw the NULL creates semantic issues for the transpilation. After, detecting all this edge cases let's make a following PR that contains the robust transpilation.

[geooo109]

There are cases that the transpilation results into different semantics.

clickhouse (input):
SELECT arrayMap(NULL, [1, 2]) AS res

Query id: 0db4a81e-03b1-4253-8593-6a73112a447f

   ┌─res──┐
1. │ ᴺᵁᴸᴸ │
   └──────┘

1 row in set. Elapsed: 0.002 sec. 

transpiled duckdb (output):
memory D SELECT LIST_TRANSFORM([1,2], NULL) AS res;
Binder Error:
Invalid lambda expression!

transpiled spark 4 (output):
spark-sql (default)> SELECT TRANSFORM(array(1,2), NULL);
[null,null]
Time taken: 0.051 seconds, Fetched 1 row(s)

[geooo109]

Same as https://github.com/tobymao/sqlglot/pull/7794/changes#r3473795554 :

clickhouse (input):
SELECT arrayFilter(NULL, [1, 2, 3])

Query id: 2b70a67e-8db5-47e5-b60a-02551e878843

   ┌─arrayFilter(NULL, [1, 2, 3])─┐
1. │ ᴺᵁᴸᴸ                         │
   └──────────────────────────────┘

1 row in set. Elapsed: 0.002 sec. 

transpiled duckdb(output): 
memory D SELECT LIST_FILTER([1, 2, 3], NULL);
Binder Error:
Invalid lambda expression!

[geooo109]

Let's remove the comments here, to simplify the code.