Substring type inference missing binary case

[racevedoo]

Fully reproducible code snippet

>>> optimize(sqlglot.parse_one("select col from tbl",read='spark'),schema={"tbl": {"col": "BINARY"}}).expressions[0].type
DataType(this=Type.BINARY, nested=False)
>>> optimize(sqlglot.parse_one("select substring(col, 2, 3) as x from tbl",read='spark'),schema={"tbl": {"col": "BINARY"}}).expressions[0].type
DataType(this=Type.VARCHAR)

In the second expression, the expected output type is Type.BINARY. In spark shell, this works fine (spark 3.5.1):

scala> val df = spark.sql("select cast('thing' as binary)")
df: org.apache.spark.sql.DataFrame = [CAST(thing AS BINARY): binary]

scala> val df = spark.sql("select substring(cast('thing' as binary), 1, 3)")
df: org.apache.spark.sql.DataFrame = [substring(CAST(thing AS BINARY), 1, 3): binary]

From my understanding, the return data type for Substring is hardcoded to VARCHAR:

sqlglot/sqlglot/dialects/dialect.py

Lines 609 to 617 in fcaae87

	exp.DataType.Type.VARCHAR: {
	exp.ArrayConcat,
	exp.Concat,
	exp.ConcatWs,
	exp.DateToDateStr,
	exp.GroupConcat,
	exp.Initcap,
	exp.Lower,
	exp.Substring,

[racevedoo]

Thanks for the quick fix!