Feat/sdbql set operations - #7

… KEEP, NONE, OFFSET

Query blocks combine with UNION [ALL] / INTERSECT / EXCEPT. Either side may be
parenthesized, chains follow SQL precedence (INTERSECT binds tighter than
UNION/EXCEPT, which chain left to right), and rows compare by value — the same
equality the UNION()/INTERSECTION() array builtins use.

WITH RECURSIVE name AS (<anchor> UNION ALL <step>) iterates the step until it
stops producing rows, binding the CTE name inside the step to the previous
iteration's rows, capped at 1,000 iterations / 1M rows.

Also: RETURN DISTINCT, COLLECT ... INTO g KEEP v1, v2 (an unknown name is an
error), the NONE quantifier and its NONE(arr, x -> cond) function form, and
OFFSET standalone or as LIMIT n OFFSET m.

Nested query blocks are no longer second-class: CTE bodies, set-operation
operands, recursive anchors/steps and correlated subqueries all run through one
entry point, so their own WITH, pre-FOR LETs, SORT/LIMIT and nested set
operations apply. A CTE declared before a set operation binds in every operand
and is evaluated once. RETURN DISTINCT is applied after the pipeline, so every
fast path honours it, columnar collections included.

LimitClause.count is Option<Expression>: a standalone OFFSET has no count
rather than a sentinel maximum, which used to reach scan_values_range as an
allocation hint and panic with "capacity overflow".

Permission checks, query caching, cache invalidation, the long-running-query
gate and has_mutations() all see through set-operation operands and CTE bodies,
so a mutation cannot hide in one and be served from cache or skip a
transaction's WAL and locks.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>