player_game data model is broken

[tobert]

With the AI playing the game at 1 game/second it pushes enough writes into this table to cause compaction issues on the slow CPUs of portacluster. It's also not partitioned well, so one node gets the brunt of the writes instead of having it move around the cluster, compounding the problem.

Solution: create a new table that has a time series bucket that can be used as the primary key, then have columns of game_id => player. Queries will have to do a bit more work but it can then store data indefinitely and the TTL can be dropped as well.