Performance tuning - Write on demand using Pull and remove toUnicastPublisher

[ikhoon]

Motivation:

The conversion between fs2.Stream and Reactive Streams Publisher
is one of the bottlenecks in that benchmark of #4

Modifications:

• Write on demand instead of Publisher

Result:

• Before

  Running 30s test @ http://127.0.0.1:8080/hello
    12 threads and 400 connections
    Thread Stats   Avg      Stdev     Max   +/- Stdev
      Latency    11.76ms    2.97ms  86.66ms   89.53%
        Req/Sec     1.71k     0.94k    3.64k    52.50%
    611170 requests in 30.02s, 107.62MB read
    Socket errors: connect 155, read 170, write 0, timeout 0
  Requests/sec:  20358.15
  Transfer/sec:      3.58MB
  

• After

  Running 30s test @ http://127.0.0.1:8080/http4s/thread
    12 threads and 400 connections
    Thread Stats   Avg      Stdev     Max   +/- Stdev
      Latency     3.77ms    2.41ms 145.85ms   96.60%
      Req/Sec     5.39k     2.88k   59.63k    60.04%
    1932651 requests in 30.10s, 312.64MB read
    Socket errors: connect 155, read 168, write 0, timeout 0
  Requests/sec:  64207.03
  Transfer/sec:     10.39MB
  
  Running 30s test @ http://127.0.0.1:8080/http4s/thread
    12 threads and 400 connections
    Thread Stats   Avg      Stdev     Max   +/- Stdev
      Latency     3.92ms    1.30ms  63.05ms   95.09%
      Req/Sec     5.11k     2.48k   10.36k    61.56%
    1831547 requests in 30.03s, 296.28MB read
    Socket errors: connect 155, read 160, write 0, timeout 0
  Requests/sec:  60989.03
  Transfer/sec:      9.87MBow it will be ignored.