Made reward function with more parameters in order to mitigate getting stuck in local minima without random stages

Change mario save/load, cnn_test.py can be used to make moveis of all maps, bestrewardsfinder changed to support 1000 series marios

Added "catch up" by changing tm_globals.hyperparameters.Ws and .bs which is used to create new configs, to be the latest highest statted mario

WorkerV2/TaskMasterV3 successfully replicating results with two workers on the same machine. Might still bug out