Add tests to benchmark typechecking

[shonfeder]

@hvanz has reported that type checking is noticeably slower when using sum types. Some increased complexity is to be expected given the implementation, but we don't want it to reduce the UX.

We should probably add some benchmarking tests to catch perf regressions so we can at least be clear about the tradeoffs if we add features or changes that increase the run time of commands.