Add model architecture (nanochat/), config.json, and inference.py

Document optimizer shard / resuming pretraining

Add base checkpoint (step 2162)