[For Testing] Go faster stripes MkII - Collated changes

[benaadams]

Updated #305

Have now split into 10 PRs:
#310 MemoryPool2 Stack->Queue, Allocate returns newest
#311 Task.Run -> ThreadPool.QueueUserWorkItem
#312 Reduce GetString allocs and conversions
#315 Lockless socket output, lower allocs (4 inter dependent) commits
#316 Remove KestrelThread locks
#317 Connection - remove closure allocation
#318 Listener - remove closure allocation
#319 ListenerPrimary - remove closure allocation
#320 ListenerSecondary - reduce closure allocation
#321 Reduce delegate allocation

Can't currently performance test as the feeds aren't matching source

[benaadams]

Baseline coreclr x64, third run v1.0.0-rc2-15944 77% CPU

~/wrk$ wrk -c 256 -d 30 -t 32 ... -s ./scripts/pipeline.lua -- 16
Running 30s test @ ...
32 threads and 256 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 4.54ms 8.18ms 135.70ms 96.37%
Req/Sec 21.44k 3.10k 55.99k 90.54%
20515710 requests in 30.10s, 2.52GB read
Requests/sec: 681589.56
Transfer/sec: 85.80MB

[benaadams]

Closing for now until have some numbers; still needs work

[benaadams]

Broke 1M but still higher latency means its worse at 256

~/wrk$ wrk -c 1024 -d 30 -t 32 ... -s ./scripts/pipeline.lua -- 16
Running 30s test @ ...
32 threads and 1024 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 17.60ms 14.74ms 264.28ms 67.36%
Req/Sec 32.41k 4.61k 54.43k 70.76%
30130208 requests in 30.08s, 3.70GB read
Socket errors: connect 35, read 0, write 0, timeout 0
Requests/sec: 1001566.79
Transfer/sec: 126.08MB