Bump sync_stream_write benchmark payload to 64MB

[Kludex]

Summary

• The 4MB payload in test_bench_sync_stream_write_large wasn't forcing enough partial sends on the CodSpeed Macro Runner to expose the buffer-slicing cost that perf: use memoryview in write() to avoid copies #954 optimizes.
• Bumping to 64MB guarantees many partial-send iterations regardless of how the kernel sizes SO_SNDBUF, which is the path the memoryview optimization targets.

Test plan

• CodSpeed walltime run completes on this PR
• Re-run perf: use memoryview in write() to avoid copies #954 with main updated to this commit and confirm the memoryview change now shows a measurable improvement on test_bench_sync_stream_write_large

AI Disclaimer

This PR was developed with the assistance of either Claude or Codex. I've reviewed and verified the changes.

[codspeed-hq]

Merging this PR will degrade performance by 95.27%

❌ 1 regressed benchmark
✅ 6 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	test_bench_sync_stream_write_large	11.5 ms	243.1 ms	-95.27%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing bump-bench-payload-64mb (1ab0bb9) with main (4fecf95)}