Feat(duckdb): handle transpilation into DuckDB from ByteString type

[fivetran-felixhuang]

At the moment, the string in a ByteString is tranpiled to a string with the escape syntax e'...'. However, DuckDB has limited support for e'...'

We need to handle the escape sequences in the ByteString input correctly, while also making sure the resulting DuckDB query produces the same result as the original BigQuery query

To handle escape sequences, we can use the ::blob operator, and to handle other possible utf-8 characters, we can use the encode() function in DuckDB. We have to use ::blob and encode() for different input segments.

For one, ::blob doesn't handle utf-8 characters after the first 256 ones (such as 数).

Also, while encode can handle escape sequences, it treats them as string literals instead of actual bytes, so the resulting query can produce different values. For example, MD5(b"Mixed\x00Texÿt") in BQ and base64(UNHEX(MD5(ENCODE('Mixed\x00Texÿt')))) in DuckDB produce different outputs

The strategy here is to handle escape sequences and other segments separately and differently, and concatenate them as the output

Examples of BQ to DuckDB

MD5(b"Mixed\x00\x00Texÿt") -> UNHEX(MD5(ENCODE('Mixed') || '\x00\x00'::BLOB || ENCODE('Texÿt')))
so "Mixed\x00\x00Texÿt" is broken into ['Mixed', '\x00\x00', 'Texÿt']

MD5(b'\x00ÿ\x00') -> UNHEX(MD5('\x00'::BLOB || ENCODE('ÿ') || '\x00'::BLOB))
MD5(b'ÿ数据') -> UNHEX(MD5(ENCODE('ÿ数据')))
MD5(B"Hello World") -> UNHEX(MD5(ENCODE('Hello World')))

[georgesittas]

@fivetran-felixhuang let's discuss this on Monday. This approach feels too complicated.