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

refactor(mediapipe): enhance MediaPipePoseBackend to prefer tasks on GPU and improve delegate handling

feat(coach): disable verifier by default

docs: add coach summary training report