Optimize assembly tracing

[bsideup]

On Java 8, getStackTraceDepth + getStackTraceElement are much faster than new Exception().getStackTrace().
There are no such methods in Java 9, but StackWalker API, which is a bit slower than native ones, but also outperforms getStackTrace().

If I run the provided benchmark on my machine (Mac OS X, 2,2 GHz Intel Core i7), I get the following numbers:
Reactor core 3.2.3.RELEASE:

Benchmark                                  (stackSize)  Mode  Cnt       Score       Error  Units
AssemblyTraceBenchmark.withTracingOnJDK11           10  avgt    5  296039.399 ± 27838.411  ns/op
AssemblyTraceBenchmark.withTracingOnJDK11           40  avgt    5  513306.643 ± 11894.858  ns/op
AssemblyTraceBenchmark.withTracingOnJDK11           80  avgt    5  800795.784 ± 48429.266  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            10  avgt    5  203167.404 ±  2892.552  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            40  avgt    5  385853.825 ± 12205.563  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            80  avgt    5  617802.999 ± 12685.994  ns/op
AssemblyTraceBenchmark.withoutTracing               10  avgt    5    3180,662 ±   187,830  ns/op
AssemblyTraceBenchmark.withoutTracing               40  avgt    5    4745,460 ±   121,963  ns/op
AssemblyTraceBenchmark.withoutTracing               80  avgt    5    6190,435 ±   431,500  ns/op

This branch:

Benchmark                                  (stackSize)  Mode  Cnt      Score      Error  Units
AssemblyTraceBenchmark.withTracingOnJDK11           10  avgt    5  48189,066 ± 1246,476  ns/op
AssemblyTraceBenchmark.withTracingOnJDK11           40  avgt    5  50204,746 ± 2709,962  ns/op
AssemblyTraceBenchmark.withTracingOnJDK11           80  avgt    5  51761,306 ±  663,210  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            10  avgt    5  36983,888 ± 2164,248  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            40  avgt    5  46101,357 ± 1066,614  ns/op
AssemblyTraceBenchmark.withTracingOnJDK8            80  avgt    5  60321,146 ± 2099,422  ns/op
AssemblyTraceBenchmark.withoutTracing               10  avgt    5   3052,667 ±   56,492  ns/op
AssemblyTraceBenchmark.withoutTracing               40  avgt    5   4629,085 ±  181,765  ns/op
AssemblyTraceBenchmark.withoutTracing               80  avgt    5   6045,648 ±  169,674  ns/op

As you can see, it is 4-10x times faster and much less dependant on the stack size. Also, the error is lower.

Open questions

• nasty sun.misc.* hack. I haven't figured out how to do it with JDK 9+ and -release 8 (see Remove sun.misc.* stubs #1469)

[codecov-io]

Codecov Report

Merging #1466 into master will increase coverage by 0.2%.
The diff coverage is 78.88%.

@@             Coverage Diff             @@
##             master    #1466     +/-   ##
===========================================
+ Coverage     84.27%   84.48%   +0.2%     
- Complexity     3895     4216    +321     
===========================================
  Files           358      359      +1     
  Lines         29435    30528   +1093     
  Branches       5464     5671    +207     
===========================================
+ Hits          24806    25791    +985     
- Misses         3030     3105     +75     
- Partials       1599     1632     +33

Impacted Files	Coverage Δ	Complexity Δ
...in/java/reactor/core/publisher/MonoOnAssembly.java	92.3% <100%> (-2.14%)	7 <0> (-2)
...reactor/core/publisher/FluxCallableOnAssembly.java	80% <100%> (ø)	6 <0> (ø)	⬇️
...ore/src/main/java/reactor/core/publisher/Flux.java	98.97% <100%> (+0.23%)	711 <2> (+201)	⬆️
...re/src/main/java/reactor/core/publisher/Hooks.java	96.31% <100%> (+1.08%)	41 <7> (+6)	⬆️
...in/java/reactor/core/publisher/FluxOnAssembly.java	91.76% <100%> (+0.85%)	16 <0> (-4)	⬇️
...ore/src/main/java/reactor/core/publisher/Mono.java	94.76% <100%> (+0.95%)	354 <2> (+92)	⬆️
...reactor/core/publisher/MonoCallableOnAssembly.java	72.22% <100%> (ø)	8 <0> (ø)	⬇️
...main/java/reactor/core/publisher/SignalLogger.java	74.45% <100%> (ø)	70 <0> (ø)	⬇️
...ctor/core/publisher/ConnectableFluxOnAssembly.java	85.71% <100%> (ø)	8 <1> (ø)	⬇️
...reactor/core/publisher/ParallelFluxOnAssembly.java	69.23% <100%> (-8.55%)	10 <1> (-3)
... and 31 more

---

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 4f9e176...07a2787. Read the comment docs.

[smaldini]

Maybe not Tracer to keep that naming space for tracing (aka zipkin) features

[bsideup]

I removed the class, not needed anymore. The content moved to Traces where it belongs :)