bug: Stream map defined using record or _ syntax fails to look up correct data type

[dluo-sig]

Another similar issue in CustomStreamMap:

                existing_schema: dict = (
                    # Use transformed schema if available
                    transformed_schema["properties"].get(prop_key, {})
                    # ...or original schema for passthrough
                    or self.raw_schema["properties"].get(prop_def, {})
                )

If the stream map is defined using _['columnName'] or record['columnName'], the type mapping falls back to string because it doesn't look at the underlying column name, but rather treats the entire expression as the column that it's searching for. prop_def should check for that specification and pass through the appropriate column name.

Originally posted by @dluo-sig in #1695 (comment)

[dluo-sig]

Also, another thing to note is that when renaming a column and assigning the original to __NULL__, the null currently must be defined second in order to pick up the data type here. Otherwise, it gets popped from the schema before the type can be grabbed. One potential solution is to order the stream maps, but not sure if this might have some other downstream effect.

for prop_key, prop_def in sorted(stream_map.items(), key=lambda item: item[1] == "__NULL__"):

[dluo-sig]

Here is also the transformation I applied to prop_def to get the actual column name. Again, this has not been extensively tested yet.

# Extract the property name from prop_def if it matches the pattern _['ColumnName'] or record['ColumnName']
def extract_property_name(prop_def):
    match = re.match(r"[_record]+\['(.+?)'\]", prop_def)
    return match.group(1) if match else prop_def