Adds PipeWriterAdapter

[jkotalik]

Half of https://github.com/aspnet/AspNetCore/issues/3966.

Few questions:

• What should we do with OnReaderCompleted? If we implement it, it means that we would probably need to create a PipeReaderAdapter (or somehow make a shared object).
• I think I need to synchronize FlushAsync and CancelPendingFlush

Tests

• I took tests from https://github.com/dotnet/corefx/blob/master/src/System.IO.Pipelines/tests/PipeWriterTests.cs for all writer behavior.
• It will be difficult to directly port any tests that use both PipeReader and PipeWriter with the adapter implementations. This is because having a PipeReader/PipeWriter with a shared backing stream will not work because it would require seeking/setting positions to make it work.
• I also added tests for cancellation

Performance

Current performance numbers:

Method	Mean	Error	StdDev	Median	Op/s	Gen 0/1k Op	Gen 1/1k Op	Gen 2/1k Op	Allocated Memory/Op
WriteHelloWorld	6.280 us	0.1119 us	1.033 us	5.965 us	159,246.75	-	-	-	4.72 KB
WriteHelloWorldLargeNumberOfWrites	2,515.595 us	23.2819 us	222.184 us	2,512.710 us	397.52	-	-	-	4256.74 KB
WriteHelloWorldLargeWrite	22.644 us	0.3660 us	3.410 us	22.165 us	44,162.09	-	-	-	117.45 KB
WriteHelloWorldLargeNumberOfLargeWrites	16,120.442 us	70.6716 us	652.178 us	15,988.595 us	62.03	1000.0000	1000.0000	1000.0000	28904.27 KB

[jkotalik]

Better perf numbers.

BenchmarkDotNet=v0.10.13, OS=Windows 10.0.17134
Intel Xeon CPU E5-1650 v4 3.60GHz, 1 CPU, 12 logical cores and 6 physical cores
.NET Core SDK=2.2.100-preview2-009404
  [Host]     : .NET Core 2.2.0-preview2-26905-02 (CoreCLR 4.6.26905.03, CoreFX 4.6.26905.02), 64bit RyuJIT
  Job-MVJLBU : .NET Core 2.2.0-preview2-26905-02 (CoreCLR 4.6.26905.03, CoreFX 4.6.26905.02), 64bit RyuJIT

Runtime=Core  Server=True  Toolchain=.NET Core 2.2  
RunStrategy=Throughput  

Method	Mean	Error	StdDev	Op/s	Allocated
WriteHelloWorld	149.7 ns	2.952 ns	4.327 ns	6,679,947.5	32 B
WriteHelloWorldLargeWrite	60,649.7 ns	812.198 ns	759.731 ns	16,488.1	100002 B

[jkotalik]

Perf numbers:

Using Memory and MemoryPool:

Method	Mean	Error	StdDev	Op/s	Allocated
WriteHelloWorld	200.1 ns	3.846 ns	4.579 ns	4,998,578.2	32 B
WriteHelloWorldLargeWrite	61,825.2 ns	1,081.980 ns	959.147 ns	16,174.6	100003 B

Using byte[] and ArrayPool:

Method	Mean	Error	StdDev	Op/s	Allocated
WriteHelloWorld	149.7 ns	2.952 ns	4.327 ns	6,679,947.5	32 B
WriteHelloWorldLargeWrite	60,649.7 ns	812.198 ns	759.731 ns	16,488.1	100002 B

[davidfowl]

https://github.com/aspnet/HttpAbstractions/pull/1065#issuecomment-438898235

These numbers don't sit well with me. I don't see how the allocated bytes can be that similar.

[pakrym]

What are the latest perf results?

[jkotalik]

Removed the MemoryStream and made an underlying noop stream instead.

Method	Mean	Error	StdDev	Op/s	Allocated
WriteHelloWorld	179.2 ns	3.603 ns	6.404 ns	5,580,116.0	0 B
WriteHelloWorldLargeWrite	2,029.8 ns	40.592 ns	94.883 ns	492,654.3	0 B