Build simple profiling tooling

[braxtonmckee]

For instance, a flag could be set that adds instrumentation to every compiled TP function counting every time we enter a function and every outgoing callsite. At each point, we'd end up with a new global variable (we'd have to add something to GlobalVariableDefinition to model this) containing a raw pointer to an integer and we'd just increment it nakedly every time we execute.

We could then walk back over the stats and figure out which function call sites were being used.

An alternative would be for each thread to have its own 'event pointer ring buffer' and we'd write events into the ring buffer, which could conceivably be richer data, since it would tell us the ordering of events and also let us use sampling to determine where a program was spending its time.

Either way, the goal would be to make a first step towards being able to tell where a TP program is spending its time.