Fix!: REGEXP_EXTRACT position arg overflow

[treysp]

Context

REGEXP_EXTRACT extracts substrings based on a regex pattern (and is called REGEXP_SUBSTR in multiple dialects).

In some dialects, REGEXP_EXTRACT has a position arg that specifies an index value for where in the string the matching should begin. This effectively means that a SUBSTR operation occurs before the regex pattern match.

DuckDB does not support the position arg, so we match the source semantics with a SUBSTR call inside the REGEXP_EXTRACT call.

Problem

What should happen when the position index value is larger than the length of the string?

In SUBSTR, the semantics are that it returns an empty string '', but in REGEXP_EXTRACT all dialects other than Redshift return NULL.

Therefore, for those source dialects we must wrap the DuckDB SUBSTR call with NULLIF(SUBSTR(arg, pos), '').

Wrinkle

REGEXP_EXTRACT always returns a single match or capture group (either the first by default or from a user-specified arg). In contrast, REGEXP_EXTRACT_ALL returns all matches or capture groups in an ARRAY.

When position overflows in a REGEXP_EXTRACT_ALL call, an empty array is returned that contains neither empty strings nor NULL values.

closes #6442