weiweihook/RLPlanner ? reverse-engineered prompt

Build me a small research style tool called RLPlanner that can take a 2.5D chiplet system from a config file and find a good floorplan using reinforcement learning.

I want it to place chiplets on an interposer, try to reduce both total wirelength and the hottest temperature, and show the final placement plus training plots and result summaries. It should support the kinds of example systems mentioned in the README, like multi GPU, CPU DRAM, Ascend910, and a few synthetic cases, with chiplet sizes, power, and connection matrix coming from the config.

Also include a fast thermal estimation method so it does not need to run a slow solver every time, and make it easy to compare against a slower baseline if needed. Please set it up so I can run experiments from the command line, save outputs into a results folder, and use the existing config files in the repo. If anything is unclear, look up the current docs online if you need to.