[BUG] legacy cast of a struct column to string with a single nested null column yields null instead of '[]'

[gerashegalov]

Describe the bug
In Spark 3.0.x and Spark 3.1+ with the legacy mode, casting a struct with a single nested null-valued field produces a null string instead of []

Steps/Code to reproduce bug

import pyspark.sql.functions as F
import pyspark.sql.types as T

# for Spark 3.1+
spark.conf.set('spark.sql.legacy.castComplexTypesToString.enabled', True)

data = [ (('gera',),), ((None,),), (None,) ]
df = spark.createDataFrame(data)
df.select(df._1.cast(T.StringType())).collect()

Broken output:

[Row(_1='[gera]'), Row(_1=None), Row(_1=None)]

Expected behavior
Correct output:

[Row(_1='[gera]'), Row(_1='[]'), Row(_1=None)]

Environment details (please complete the following information)
Reproduces in a local REPL

Additional context
CI test failures in #2274

[gerashegalov]

Non-legacy cast is correct.

[gerashegalov]

related to #1604