fix: unified query, drop pipeline
perf: GROUP BY MATCH deep-hop now 3x faster via frontier BFS collect_final_dest_counts(): merged frontier HashMap<hash,count> through  hop chain, no per-path RawPath materialisation. 81k paths → 34 entries.  Ultra-fast path in execute_match_agg for final-var-only GROUP BY.