[kvlonge]

Stacked on #98 (fix/d215-lag-lead); review this diff against that branch.

What changes

REGEXP_SUBSTR now converts when its position, occurrence or group is computed (a column or expression), not only a literal (regexp_substr_computed). It follows A's receipts (regexp-substr.column-arguments, already on #98):

• A NULL argument gives NULL.
• A position or occurrence below 1, or a negative group, raises 100050 (SQLSTATE 22023). That is a new known Snowflake error of kind InvalidArgument: INVALID_PARAMETER_VALUE / REGEXP_PARAMETER_FAILURES in sqlquack-core.
• A start past the end gives NULL, and ^ anchors at the start position.
• Group 0 is the whole match, and a group past the pattern's last gives NULL.
• A word boundary after a start position is refused, since no recording shows it. Parameters still need a text literal.

Evidence

• regexp-substr t002, t003, t007, t008 and t010, plus column-arguments t001–t010, now match. t009 (parameters from a column) stays refused.
• verify-recorded: 11,606 cases, 7,671 passed, 0 failed, 220 of 220 verified.
• Translation baseline against D215: LAG/LEAD negative offsets, POSITION start, SPLIT_PART column delimiter, GENERATOR constants #98's (4ce4490): 0 regressions. 15 statements moved from refused to matching, with no form changes.

The D94 gate runs on the landing head after #98 lands.

Review fix (f8a7eac)

On the computed path, a pattern that matches empty text now raises regexp_pattern_syntax, as REGEXP_COUNT and REGEXP_INSTR do. The receipts cover a start at or past the end only for patterns that can't match empty text. All 15 recorded cases still match, and the other 12 REGEXP_SUBSTR case files are unchanged (178 cases, 0 failed).

Follow-ups

• The computed path declares the regexp_pattern_syntax divergence for every pattern. Since the pattern is a literal, conversion could decide whether it matches empty text and skip the divergence when it can't.
• A receipt for an empty-matching pattern at p = length + 1 would let the raise be lifted.

[kvlonge]

Closing: SQLQuack lands through local-gate trains, not PR merges (D94), so open PRs only add clutter (D628/D630). The branch is kept. Any work here that isn't on main yet is tracked by the train owner and lands through a train.