IssueTracker

Currently, the primary way to configure benchmark behavior (looping, profiling, etc) is to set instrumentation arguments.

This is nice for a few reasons:
* Change properties without rebuilding
* Set easily from CLI

But it has several problems:
* Not discoverable
* Doesn't make sense for properties you want to leave on, requires external configuration
* Only configurable module-wide
* Doesn't scale to complex inter-related properties

We also started internally experimenting with constructor properties, like configuring tracing for microbench, but this the similar (smaller) problem that it's only configurable class-wide.

Profiling and level of tracing would both be hugely useful to configure in code, but we should define a scalable configuration API for this.