Add HTTP/2 HPack static compression

[JamesNK]

Fixes #4716

I haven't benchmarked anything yet but response HEADERS frames in unit tests are expectedly smaller.

There are a lot of test failures, but that is because I haven't updated the expected length for HEADERS. Will deal with those once we're happy with the approach.

[benaadams]

If you are adding an enum KnownHeaderType change it to flags and then use it for the constants?

- if ((_bits & 0x1L) != 0)
+ if ((_bits & KnownHeaderType.CacheControl) != 0)
  {
      _current = new KeyValuePair<string, StringValues>(HeaderNames.CacheControl, _collection._headers._CacheControl);
      _currentKnownType = KnownHeaderType.CacheControl;
      _next = 1;
      return true;
  }

[Tratcher]

Any luck cleaning up HPackEncoder.BeginEncode?

Any influence on benchmarks?

I was going to ask about static value encoding, but status is the only response header with values in the static table isn't it?

Not your problem, but still not pleased about the duplication here between the client and the server.

[Tratcher]

I had envisioned this method and the enum being in the shared code, but I can understand why you would want to stop here. The client has almost this exact code for request headers but it will take some more work to deduplicate.

[JamesNK]

We're at least now using the same static helper methods.

I'm not sure where the client would use a method like EncodeHeader. Many of their headers are written explicitly, and their unknown headers are written here - https://github.com/dotnet/runtime/blob/master/src/libraries/System.Net.Http/src/System/Net/Http/SocketsHttpHandler/Http2Connection.cs#L1021

I'm not sure how their header buffer handles splitting content across multiple HEADER frames.

Is there somewhere else in the client that I'm missing?

[Tratcher]

I meant that translations like HPackHeaderMapping.GetResponseHeaderStaticTableId would be done inside the HPack implementation.

[JamesNK]

If you are adding an enum KnownHeaderType change it to flags and then use it for the constants?

Maybe. KnownHeaderType enum has all the known request and response values in it. It might be too large for a long flags enum.

We might end up using KnownHeaderType as an index in an array of header metadata. That metadata would include the static compression ID, pre-encoded header names, etc. A flag enum wouldn't work in that situation.

[Tratcher]

Mainly waiting on HPackEncoder.BeginEncode cleanup and benchmarks.

[JamesNK]

This is copy and paste (with some modifications to remove some internal type dependencies) but that appears to match the other code in http2cat which is copied from Http2TestBase

[JamesNK]

State gone! @scalablecory @Tratcher