Expose response PipeWriter in Kestrel

[jkotalik]

For #4351 and does many other things too that are required for this change.

• Exposes the PipeWriter as directly as possible from Kestrel (for both Http1 and Http2)
  • Pay careful attention to Http1 chunking logic as that actually required work.
• Changes most tests to start using the PipeWriter

Things left to do:

• Add tests for setting response body/ response pipe and how they override each other.
• Perf tests (micro benchmarks and before/after with benchmarks).

Have fun everyone 😄

[jkotalik]

Some feedback for myself and things to figure out.

[jkotalik]

Looks like swapping WriteAsync broke stuff too. I think they are issues with StreamPipeWriter so I'll investigate and create a separate PR.

[JunTaoLuo]

👀

[sebastienros]

/AzurePipelines run AspNetCore-benchmarking-pr

[azure-pipelines]

Successfully queued 1 pipeline(s).

[jkotalik]

Figured out why all response compression tests are failing. We are now double flushing the stream for ResponseCompression; once from StreamPipeWriter.FlushAsync, and another from BodyWrapperStream.FinishCompressionAsync (which calls Dispose), which causes extra compression frames to be written to the response.

To fix this, either we need a way to either make the StreamPipeWriter not call FlushAsync on the underlying stream and only write the buffered contents to the stream (which may straight up break scenarios), or we need to fix the tests to expect a longer length (which isn't desired as it means response compression is less optimized).

[Tratcher]

Go ahead and update the response compression tests (after verifying the data isn't corrupt and can round trip). An extra flush at the end isn't the end of the world. We do want to discourage frequent flushes though.

[jkotalik]

🆙 📅 . There is still some crust around the new Disposal logic I added, but I wanted to make sure it's a good idea before cleaning up.

[jkotalik]

Bad merge. Fixing.

[benaadams]

1 error on AspNetCore-helix-test but only on Centos.7.Amd64.Open

[xUnit.net 00:00:02.41]     Microsoft.AspNetCore.Components.Test.RendererTest.CanTriggerEventHandlerDisposedInEarlierPendingBatchAsync [FAIL]
Failed   Microsoft.AspNetCore.Components.Test.RendererTest.CanTriggerEventHandlerDisposedInEarlierPendingBatchAsync
Error Message:
 Assert.Throws() Failure
Expected: typeof(System.ArgumentException)
Actual:   (No exception was thrown)
Stack Trace:
   at Microsoft.AspNetCore.Components.Test.RendererTest.CanTriggerEventHandlerDisposedInEarlierPendingBatchAsync() 
in /_/src/Components/Components/test/RendererTest.cs:line 1364
--- End of stack trace from previous location where exception was thrown ---

[jkotalik]

Fortunately Helix tests aren't required yet 😄