chore: add ninja-build and build-essential to package requirements; update README and documentation to reflect changes in model configuration and local Transformers usage

chore: update base model to NVIDIA-Nemotron-3-Nano-4B-BF16 across configurations and documentation; adjust max sequence length to 2048; upgrade GitHub Actions cache to v5

feat: expand coach summary training pipeline

fix: align coach summary Modal training with current TRL API

feat: add Modal pipeline for coach summary training