Revisit benchmarker for HTTP benchmarks

While analyzing the performance improvement of Node.js 20, I faced an unexpected behavior, see:

$ ./node18 benchmark/run.js --filter "simple.js" http

http/simple.js
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="bytes" benchmarker="test-double-http": 57,011
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="buffer" benchmarker="test-double-http": 54,078

$ ./node20 benchmark/run.js --filter "simple.js" http

http/simple.js
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="bytes" benchmarker="test-double-http": 127,857
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="buffer" benchmarker="test-double-http": 116,360

A noticeable performance improvement (from 57,011 to 127,857), but if you switch the benchmarker method, this performance drops:

$ npm i -g autocannon

added 48 packages in 5s

5 packages are looking for funding
  run `npm fund` for details
$ ./node18 benchmark/run.js --filter "simple.js" http

http/simple.js
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="bytes" benchmarker="autocannon": 42,648.73
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="buffer" benchmarker="autocannon": 43,250.91
$ ./node20 benchmark/run.js --filter "simple.js" http

http/simple.js
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="bytes" benchmarker="autocannon": 46,552.73
http/simple.js duration=10 chunkedEnc=0 c=100 chunks=4 len=1024 type="buffer" benchmarker="autocannon": 46,477.1

This indicates that the previous benchmarking method (test-double-http) had limitations or errors. It might be due to the keep-alive turned on by default after Node.js 19?

Start with benchmark/run.js and the http/simple.js benchmark, then reproduce the Node.js 18 and 20 results using both test-double-http and autocannon. Investigate whether differing keep-alive behavior explains the discrepancy; done means the benchmarker's limitation or error is identified and the benchmark results or method are corrected accordingly.