Add Microbenchmarking

[marcotc]

This PR adds microbenchmarks to core components of ddtrace.
Currently we test:

• Trace creation (tracer.trace)
• Writer trace recording (writer.write)
• Baseline benchmark overhead: tests an empty method call, to validate if the benchmark environment is stable.

All test cases are measure for execution time, memory usage and GC impact.

The output looks like this (for testing Datadog::Tracer nested traces in this example):

Test:Microbenchmark Datadog::Tracer nested traces memory
Calculating -------------------------------------
                   1   253.544k memsize (    91.440k retained)
                         2.830k objects (   704.000  retained)
                         4.000  strings (     1.000  retained)
                  10     1.811M memsize (   862.280k retained)
                        14.488k objects (     5.661k retained)
                         3.000  strings (     0.000  retained)
                 100    17.340M memsize (     8.508M retained)
                       131.485k objects (    55.171k retained)
                         3.000  strings (     0.000  retained)

Comparison:
                   1:     253544 allocated
                  10:    1810560 allocated - 7.14x more
                 100:   17339976 allocated - 68.39x more

Test:Microbenchmark Datadog::Tracer nested traces timing
Warming up --------------------------------------
                   1     4.176k i/100ms
                  10     1.342k i/100ms
                 100   192.000  i/100ms
Calculating -------------------------------------
                   1     39.479k (±13.5%) i/s -     41.760k in   1.080583s
                  10      9.552k (±24.4%) i/s -     10.736k in   1.202755s
                 100      1.199k (±21.5%) i/s -      1.152k in   1.006056s

Comparison:
                   1:    39479.1 i/s
                  10:     9552.2 i/s - 4.13x  (± 0.00) slower
                 100:     1198.8 i/s - 32.93x  (± 0.00) slower

There's also a detailed, human-readable memory report under detailed report tests, which reports on every memory allocation site, like this:

allocated memory by location
-----------------------------------
     24000  lib/ddtrace/sampler.rb:188
     19200  lib/ddtrace/span.rb:69
     19200  lib/ddtrace/tracer.rb:193
     19200  lib/ddtrace/tracer.rb:259
     18400  lib/ddtrace/tracer.rb:201

This should give us a good starting point to measure small units of the tracer for performance impact and objectively measure future improvements.