feat(duckdb): Add transpilation support for COLLATION function

[fivetran-amrutabhimsenayachit]

DuckDB handles collation at the column-definition level via the COLLATE clause, not as a queryable runtime function. There is no DuckDB equivalent that returns the collation specification of an expression.

Fix:
E.g:

Collation(
  this = Collate(
    this = Literal('abc'),
    expression = Literal('en-ci')   ← the spec is RIGHT HERE
  )
)

If there's an explicit COLLATE 'spec' attached and the spec is non-empty, the answer is already known at transpile time — so just write that spec as a plain string literal.
Otherwise, there is no way to know the collation, so write NULL, which is exactly what Snowflake returns for uncollated expressions.

 python3 << 'EOF'
import sqlglot
q = "SELECT COLLATION('abc') AS lit_no_coll, COLLATION(NULL) AS null_input, COLLATION('abc' COLLATE 'en-ci') AS lit_en_ci, COLLATION('abc' COLLATE 'de-ci-pi') AS lit_complex, COLLATION('abc' COLLATE 'utf8') AS lit_utf8, COLLATION('abc' COLLATE '') AS lit_empty"
print(sqlglot.transpile(q, read='snowflake', write='duckdb')[0])
EOF
-->
COLLATION function is not supported by DuckDB
COLLATION function is not supported by DuckDB
COLLATION function is not supported by DuckDB

"SELECT NULL AS lit_no_coll, NULL AS null_input, 'en-ci' AS lit_en_ci, 'de-ci-pi' AS lit_complex, 'utf8' AS lit_utf8, NULL AS lit_empty"
│ lit_no_coll │ null_input │ lit_en_ci │ lit_complex │ lit_utf8 │ lit_empty │
│    int32    │   int32    │  varchar  │   varchar   │ varchar  │   int32   │
├─────────────┼────────────┼───────────┼─────────────┼──────────┼───────────┤
│        NULL │       NULL │ en-ci     │ de-ci-pi    │ utf8     │      NULL │
└─────────────┴────────────┴───────────┴─────────────┴──────────┴───────────┘

[georgesittas]

@fivetran-amrutabhimsenayachit I don't think we should handle Collate as a special case. It's similar to why we don't want to handle NULL values explicitly. I wouldn't expect COLLATION(x COLLATE y) to show up in the wild, because that trivially evalutes to y.

I think we only want to call unsupported and do fallback_function_sql, not NULL.

[github-actions]

SQLGlot Integration Test Results

Comparing:

• this branch (sqlglot:RD-1147700-collation, sqlglot version: RD-1147700-collation)
• baseline (main, sqlglot version: 0.0.1.dev1)

By Dialect

dialect	main	sqlglot:RD-1147700-collation	transitions	links
bigquery -> bigquery	22921/22926 passed (100.0%)	21255/21255 passed (100.0%)	No change	full result / delta
bigquery -> duckdb	1300/1670 passed (77.8%)	0/0 passed (0.0%)	No change	full result / delta
duckdb -> duckdb	2425/2425 passed (100.0%)	2425/2425 passed (100.0%)	No change	full result / delta
snowflake -> duckdb	1515/2674 passed (56.7%)	0/0 passed (0.0%)	No change	full result / delta
snowflake -> snowflake	65923/65923 passed (100.0%)	63028/63028 passed (100.0%)	No change	full result / delta
databricks -> databricks	1370/1370 passed (100.0%)	1370/1370 passed (100.0%)	No change	full result / delta
postgres -> postgres	6042/6042 passed (100.0%)	6042/6042 passed (100.0%)	No change	full result / delta
redshift -> redshift	7101/7101 passed (100.0%)	7101/7101 passed (100.0%)	No change	full result / delta

Overall

main: 110131 total, 108597 passed (pass rate: 98.6%), sqlglot version: 0.0.1.dev1

sqlglot:RD-1147700-collation: 101221 total, 101221 passed (pass rate: 100.0%), sqlglot version: RD-1147700-collation

Transitions:
No change

✅ 10 test(s) passed