Use API to create a benchmark suite

To better optimize performance, we will need a reference workload, which is accurate. [With the API](https://docs.pyboy.dk), it will be easy to make a frame-accurate benchmark.

A test could load up a game (see [game-specific wrapper](https://github.com/Baekalfen/PyBoy/wiki/Example-Super-Mario-Land)) and interact with the game using button presses, and waiting for monsters, points, etc. and take action.

Each workload could run for a minute of simulated game time or more. But should be deterministic, even if frame-timing changes slightly. Manipulating the RNG might be needed.

Start with the PyBoy API documentation and the game-specific wrapper example for Super Mario Land. Define a workload that loads a game, performs button presses and waits for in-game events, and runs for at least a minute of simulated time. It is done when the benchmark is frame-accurate and deterministic despite small frame-timing changes.

Written by the indexing model from the issue text.