Move the performance-tests into separate units

[ghost]

latest results are too good:

[info] 1 http zio = 666ms / req
[info] 10 http zio = 74ms / req
[info] 1 https zio = 237ms / req
[info] 10 https zio = 64ms / req

This is because the different runs test-requests reuse earlier instantiated stuff.

Reproduce

• run the speetests
• disable the http test, run again
• disable the netty tests, run again

Workaround

• disable the http tests for now
• disable solo-netty-tests for now
• in general: enable only one test

Solution

• semi: flush/destroy everything befor each test (ZIO clients, scopes, netty)
• final: move each performance-test (curl, java-http, netty, zio-http, zio-https, ...) into its own unit (file/class/object)
  • instantiate each performance-test from a central runner (or individually via manual run)

[jamesward]

It is curious that stuff gets reused despite providing Client.default each time - maybe some state that gets initialized there. Perhaps there is another way to initialize new state each time. Worth investigating instead of turning parts of selectively.

[ghost]

Even netty is reused, independent of zio framework.

yes, this needs investigation/documentation, but for now (with the limited bounty/time budget), the workaround to enable just one test does it

https://github.com/abebeos/zio-http-client-perf/blob/main/src/main/scala/App.scala

Final solution to keep speedtests in different files would be more elegant though.

Up to you to apply it on your code here.

[ghost]

And as a sidenote: the problem is mostly with the development-mentality of the zio project. Things are not strict enough, e.g. the speedtests should be already there, right within the CI, ideally with regression-alarms and comparative-alarms (e.g. "if zio is over 50% slower that java-http, alarm").

Anyways, me out at this point, I do not expect to get even an ok for a bounty claim (voices: "where is your PR?").