fix: spark crc32 custom nullability

[watanaberin]

Which issue does this PR close?

• Closes #19157

Rationale for this change

The crc32 UDF was using the default return_type implementation which does not preserve nullability information Spark CRC32

• Only returns the data type (Int64)
• Doesn't consider nullability of inputs
• Would always mark output as non-nullable

What changes are included in this PR?

• Implemented return_field_from_args: Creates a field with Int64 type and correctly propagates nullability from input fields and scalar arguments
• Updated return_type: Now returns an error directing users to use return_field_from_args instead
• Added necessary imports: Field, FieldRef, and ReturnFieldArgs to support the new implementation
• Added comprehensive nullability tests: Verifies that nullable inputs, non-nullable inputs, and null scalar literals are handled correctly

Are these changes tested?

• Non-nullable Binary input produces non-nullable Int64 output
• Nullable Binary input produces nullable Int64 output
• Null scalar literal (e.g., crc32(NULL)) produces nullable Int64 output
• Data type is correctly set to Int64 in all cases

Are there any user-facing changes?

This is a bug fix that corrects schema metadata only, it does not change the actual computation or introduce any breaking changes to the API.