fix(cypher): preserve aggregate input with max rows

[zgqq]

Summary

• prevent max_rows from truncating rows before aggregate evaluation
• keep max_rows as a final result-row limit
• add regression coverage for COUNT with max_rows smaller than the matched input set

Verification

• regression test before fix: cypher_apply_limit FAIL: "30" != "50"
• regression test after fix: cypher_apply_limit PASS
• production build: make -j$(nproc) -f Makefile.cbm cbm
• CLI/MCP query with max_rows=10: MATCH (n) RETURN count(n) returns 685 instead of 10

The full local test runner has one environment-sensitive pre-existing failure in incr_full_index: RSS delta 2545 MiB exceeds its fixed 2048 MiB threshold. No other test failed.

Signed-off-by: zgqq zgqjava@gmail.com