Support for minimum iteration count.

At present, it is not possible to specify a minimum number of iterations for a benchmark. This is an issue when the framework decides that a given configuration for a benchmark should only be run a small number of times, fewer than we would like for the benchmark to be stable. The only way to modify the number of iterations is by explicitly specifying `Iterations(...)` for a given benchmark, or by passing `--benchmark_min_time=...` to run the benchmarks for a specific length of time.

One possible solution would be a corresponding `--benchmark_min_iterations=...` flag that could be passed at runtime, which would prevent the framework from running the benchmark for any fewer than `N` iterations, where `N` is some number by which we know that the result should be stable.