perf: avoid normal port use scheduler polling.

[qzhuyan]

since OTP21 , beam use epoll and have separate thread(s) to call the epoll.
https://www.erlang.org/blog/io-polling/

these threads can be scaled up and pollset could be scaled as well.

however, the ports will be 'upgraded' to the scheduler polling epoll set. (ERTS_POLL_USE_SCHEDULER_POLLING) if it is active and be epoll ready for 10 times.

This patch bump 10 to 999 times.

Another workaround is to have lower active_n for network ports, e.g. {active, 10}.

[qzhuyan]

Old:

3m11s connect_succ total=18062 rate=94.12/sec
3m12s pub total=14139916 rate=144886.00/sec

New:
3m41s connect_succ total=20120 rate=58.06/sec
3m42s pub total=18392634 rate=193500.54/sec

[qzhuyan]

5.10.3(core1@127.0.0.1)152> erlang:system_info(check_io).
[[{name,erts_poll},
  {primary,poll},
  {kernel_poll,false},
  {memory_size,25248},
  {total_poll_set_size,1},
  {lazy_updates,true},
  {pending_updates,0},
  {batch_updates,false},
  {concurrent_updates,false},
  {fallback,true},
  {max_fds,1024000},
  {active_fds,0},
  {poll_threads,0}],
 [{name,erts_poll},
  {primary,epoll},
  {kernel_poll,epoll},
  {memory_size,56},
  {total_poll_set_size,7},  <----------正常
  {lazy_updates,false},
  {pending_updates,0},
  {batch_updates,false},
  {concurrent_updates,true},
  {fallback,false},
  {max_fds,1024000},
  {active_fds,0},
  {poll_threads,0}],
 [{name,erts_poll},
  {primary,epoll},
  {kernel_poll,epoll},
  {memory_size,114744},
  {total_poll_set_size,1804}, <----------正常
  {lazy_updates,false},
  {pending_updates,0},
  {batch_updates,false},
  {concurrent_updates,true},
  {fallback,false},
  {max_fds,1024000},
  {active_fds,0},
  {poll_threads,4}]]