Functional tests failing due to aborted connections

[halter73]

Ex:

https://ci.appveyor.com/project/aspnetci/kestrelhttpserver/build/1.0.3648#L449
https://ci.appveyor.com/project/aspnetci/kestrelhttpserver/build/1.0.3645#L428

This reminds me of #1147, but the error is coming from BeginRead instead of EndRead.

[halter73]

I'm wondering if re-adding a line like this to wait for a client FIN/RST prior to calling uv_close is going to be the fix.

I remember removing this because it caused server shutdown to always take the full 5 seconds. Chrome and IE (but not firefox) would take forever to send a FIN back when receiving a FIN on an idle connection. For this reason, we might want to set a pretty aggressive timeout.

[halter73]

ZeroContentLengthAssumedOnNonKeepAliveRequestsWithoutContentLengthOrTransferEncodingHeader also failed on OS X now: https://travis-ci.org/aspnet/KestrelHttpServer/jobs/171495305

[halter73]

This doesn't seem to be fully fixed. SingleErrorResponseSentWhenAppSwallowsBadRequestException is still failing.

1. https://ci.appveyor.com/project/aspnetci/kestrelhttpserver/build/1.0.3704
2. Set StatusCode before disposing HttpContext (#876) #993 (comment) (From well before)

This is presumably the issue with sending extra data after the request was rejected like we used to in ZeroContentLengthAssumedOnNonKeepAliveRequestsWithoutContentLengthOrTransferEncodingHeader.