feat(duckDB): Cast inputs (BLOB → VARCHAR) for duckDB STARTS_WITH

[fivetran-amrutabhimsenayachit]

When STARTS_WITH is used with BLOB/BYTES types that are not literals (e.g., from CAST, table columns, function results), the transpiled DuckDB query fails because DuckDB's starts_with only accepts VARCHAR.

Before:

sqlglot %  bq --project_id fivetran-wild-west query --use_legacy_sql=false "SELECT STARTS_WITH(CAST('foo' AS BYTES), CAST('f' AS BYTES))"                            
+------+
| f0_  |
+------+
| true |
+------+
sqlglot % python3 -c "import sqlglot; print(sqlglot.transpile(\"SELECT STARTS_WITH(CAST('foo' AS BYTES), CAST('f' AS BYTES))\", read='bigquery', write='duckdb')[0])"
SELECT STARTS_WITH(CAST('foo' AS BLOB), CAST('f' AS BLOB))

sqlglot % duckdb -c "SELECT STARTS_WITH(CAST('foo' AS BLOB), CAST('f' AS BLOB))"                                                         
Binder Error:
No function matches the given name and argument types 'starts_with(BLOB, BLOB)'. You might need to add explicit type casts.
        Candidate functions:
        starts_with(VARCHAR, VARCHAR) -> BOOLEAN


LINE 1: SELECT STARTS_WITH(CAST('foo' AS BLOB), CAST('f' AS BLOB))

After:

sqlglot % bq --project_id fivetran-wild-west query --use_legacy_sql=false "SELECT STARTS_WITH(CAST('foo' AS BYTES), CAST('f' AS BYTES))"     
+------+
| f0_  |
+------+
| true |
+------+
sqlglot % python3 -c "import sqlglot; print(sqlglot.transpile(\"SELECT STARTS_WITH(CAST('foo' AS BYTES), CAST('f' AS BYTES))\", read='bigquery', write='duckdb')[0])"
SELECT STARTS_WITH(CAST(CAST('foo' AS BLOB) AS TEXT), CAST(CAST('f' AS BLOB) AS TEXT))

sqlglot % duckdb -c "SELECT STARTS_WITH(CAST(CAST('foo' AS BLOB) AS TEXT), CAST(CAST('f' AS BLOB) AS TEXT))"
┌───────────────────────────────────────────────────────────────────────┐
│ starts_with(CAST('foo'::BLOB AS VARCHAR), CAST('f'::BLOB AS VARCHAR)) │
│                                boolean                                │
├───────────────────────────────────────────────────────────────────────┤
│ true                                                                  │
└───────────────────────────────────────────────────────────────────────┘

[georgesittas]

We should mimic the implementation for Lower and Upper. Take a look at _case_conversion_sql.

[georgesittas]

Just one comment, looks good otherwise. Thanks!

[georgesittas]

I don't think this logic is necessary?

1. We don't have to annotate types here, since transpiling this successfully from bigquery to duckdb requires that we run type inference first, and hence we should have the types available. For duckdb -> duckdb, we don't need types– the varchar/unknown branch should suffice since the type will always be unknown and we won't cast, hence preserving the sql @ roundtrip
2. Doesn't look like we should deal with bytestring as a special case. Despite a bit verbose, doing a double-cast is probably the safer way to go.