refactor: remove Unsloth, use standard transformers + PEFT

fix: align Unsloth config with recommended GRPO settings

Merge branch 'main' of https://github.com/ademcodesproducts/OpenEnv-Inventory-Simulations

fix: pipeline-aware ordering, YoY demand signal, reward rebalancing

feat: integrate Unsloth into GRPO training pipeline

feat: full-horizon lookahead reward (365 days, <0.5ms)

feat: crash-resilient training with dataset caching and iteration resume

Remove PPO MLP agent, update README

feat: improve GRPO training logging and fix torch_dtype deprecation

Merge teammate changes, unify reward via reward.py, add PPO model

Added PPO model and reward.py

feat: improve training logging with tqdm, timings, GPU memory, ETA

Added trained PPO model + app.py UI changes for HF Spaces

Add P&L reward function, daily spoilage, stochastic lead time, and reward visualization

Add three-agent system: Claude LLM, PPO RL, and GRPO fine-tuned Qwen