HF RL Explorer

Resuming training is broken, and the default dtype won't even let me construct a config

Resuming training is broken, and the default dtype won't even let me construct a config: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). I'm trying to spin up an SAE training run with SAELens for the first time and ran into two issues that look like bugs.

Part of FineEnvs/MiMo-V2.6-RL-harbor-code.