perf: buffer accumulation in _write_query_params() and BatchMessage.send_body()

Replace per-parameter write_value(f, param) loops with list.append +
b"".join + single f.write(), reducing f.write() calls from (2*N + 1)
to 1 for N query parameters.

Applied to both:
- _QueryMessage._write_query_params() (execute/query path)
- BatchMessage.send_body() (batch path — entire message in one write)

Also removes the unnecessary super() pass-through in
ExecuteMessage._write_query_params() to eliminate MRO lookup overhead.

Benchmarks (500K iterations, best-of-5, Python 3.14, Cython .so):
  128D float32 vector (1 param):  1.25x faster
  768D float32 vector (1 param):  1.12x faster
  1536D float32 vector (1 param): 1.11x faster
  10 text columns:                1.30x faster

Includes benchmark script and 21 unit tests covering execute, query,
and batch message paths.

Author: mykaul