fix(binder): give a new aggregate its own name when an earlier clause has it

[zahariash]

This work was produced with the help of language models.

Fixes #1246.

UNWIND [1, 1, 2] AS x WITH x, count(x) AS c RETURN c, count(x) returned 1|1 twice instead of 2|1 and 1|1; sum(x), count(DISTINCT x) and count(x) + x after WITH x, count(x) + x AS c went wrong the same way. An aggregate is named after its function and arguments, so the RETURN's count(x) had the name of the expression carried in c, or of a part of it, and read that column instead of being computed.

Fix. An aggregate in a projection list whose structural name an expression in scope already has, at any depth, now gets a fresh name, as count(*) always did; the same aggregate repeated in one list shares it, so it is computed once. In WHERE and ORDER BY an aggregate keeps its structural name, so WITH x, count(x) AS c WHERE count(x) > 1 still filters on c and ORDER BY count(x) still sorts by it, and so does an aggregate inside a lambda body, which the projection does not compute.

Tests. agg/multi_query_part.test gets RepeatedAggregateAfterWithGroupKey: the four queries above, a list with count(x) twice, and the WHERE, ORDER BY and all(i IN [1] WHERE i < count(x)) cases that must keep reading c. On main the first five statements return wrong rows and the last three pass; all eight pass here. Full suite: no failures.