parse result-set rows in a single pass over the receive buffer

SELECT 1000 rows (INT, VARCHAR) takes 1006 µs with the pure-Python async
driver, against 351 µs for mariadb_c and 195 µs for asyncmy.

goal is to framing and row building in one pass over the raw buffer.
a bulk row parser that takes the receive bytearray, offset and length, parses every complete row packet in place with hoisted column metadata and no per-row slices, copies or calls, and
stops at an incomplete packet, EOF/OK, ERR or a 16 MB payload, leaving
those to the existing path.

This will improve performance