Make OpenEnv training+API judge-proof

Deploy: SOTA RL Cartesian Task and Unsloth Scripts

Harden strict (0,1) scoring boundaries across runtime and config.

Keep task scores strictly inside (0,1) in inference logs

Align inference env var handling with submission checklist.

Initial OpenEnv SQL debug environment