Count two-hop patterns grouped by a node from degrees #136

Problem

RETURN count(*) over two hops is counted from node degrees, but a grouped count is counted that way in one shape only. #113 handles RETURN m.name, count(x) over (x)-[:R]-(m)...: it counts the first hop per m in place, but the rest of the pattern still emits one row per match. Other groupings keep the join plan and build every path:

• by the root of a two-hop pattern: MATCH (a)-[:R]->(b)-[:S]->(c) RETURN a.id, count(*);
• by the middle node, RETURN b.id, count(*), which counts only the first hop per b, and only with the CSR.

Idea

• Grouped by the middle node: n1(b) × n2(b) per b, the two-hop count without its final sum.
• Grouped by the root: count the last hop per b in place (n2(b)), expand the first hop from a, and sum n2(b) per a, mirroring perf(csr): match the join path on whole-relationship queries #113 at the other end of the pattern.

Both keep the two-hop count's conditions: each WHERE condition reads one hop, and the corrections for same-type hops apply per group.

Split from #78.