Update wrk benchmark and numbers

[vkostyukov]

Upgrading the benchmark so it's now pretty much simulates TE (no inbound parsing). Also, re-run everything on my computer and updated the numbers. It's crazy, but due to recent performance optimizations in Circe and Finch, we do much better now: 95% of Finagle throughput.

/cc @spockz, @travisbrown, @rpless

[travisbrown]

Wow! And that's even without all the most recent stuff in circe.

[codecov-io]

Codecov Report

Merging #809 into master will increase coverage by 0.71%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #809      +/-   ##
==========================================
+ Coverage   79.42%   80.14%   +0.71%     
==========================================
  Files          46       46              
  Lines         700      700              
  Branches       27       27              
==========================================
+ Hits          556      561       +5     
+ Misses        144      139       -5

Impacted Files	Coverage Δ
core/src/main/scala/io/finch/Input.scala	100% <0%> (+12.5%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c07125f...b4a4cba. Read the comment docs.

[rpless]

Those numbers for Finch look great. I'm a little surprised about the Finagle ones dropping so much, though. Maybe related to the performance issue with Finagle Http/Netty 4 that you mentioned in the 0.15.0 release notes?

[vkostyukov]

@rpless Oh yeah. That's the overhead from Netty4 HTTP. It should go away in a month or two.