[Feature]: HuggingFace Hub integration via PyTorchModelHubMixin

## Problem The existing `nightmarenet/hub/` module implements custom `push_model`/`pull_model` functions but doesn't integrate with the HuggingFace Hub's standard `ModelHubMixin` pattern. This means NightmareNet models cannot be discovered on the HuggingFace Hub model browser, don't get automatic model cards, and lack `from_pretrained`/`push_to_hub` methods that the ML community expects. Sprint 7 targets HuggingFace Hub integration. ## Solution Implement `PyTorchModelHubMixin` integration: - [ ] Create `nightmarenet/hub/hf_integration.py` with `NightmareNetModelHubMixin` - [ ] Inherit from `PyTorchModelHubMixin` for automatic `save_pretrained`, `from_pretrained`, `push_to_hub` - [ ] Implement `_save_pretrained()` to serialize model weights + NightmareNet config + training metadata - [ ] Implement `_from_pretrained()` to load model + restore NightmareNet pipeline state - [ ] Auto-generate HF model card with NightmareNet-specific sections (robustness metrics, distortion config) - [ ] Register `nightmarenet` as a library on HuggingFace Hub (library tag) - [ ] Add code snippet for Hub model page ## Acceptance Criteria - `model.push_to_hub("username/model-name")` works end-to-end - `NightmareNetModel.from_pretrained("username/model-name")` loads correctly - Model card includes: training config, robustness score, distortion types used - Models appear under `nightmarenet` library filter on HuggingFace Hub - Unit tests with mocked Hub API (no real uploads in CI) - Backward compatible with existing `hub/core.py` functions ## Impact Radius | File | Why it may need attention | |------|--------------------------| | `nightmarenet/hub/hf_integration.py` | New file | | `nightmarenet/hub/core.py` | Existing push/pull - coordinate, don't duplicate | | `nightmarenet/hub/model_card.py` | Existing card generation - extend for HF format | | `nightmarenet/hub/__init__.py` | Re-export new mixin | | `pyproject.toml` | Add `huggingface-hub>=0.20.0` dependency | | `nightmarenet/training/trainer.py` | May integrate mixin for auto-push after training | | `nightmarenet/phases/evaluate.py` (L204) | Already calls `push_model` - update to use mixin | | `README.md` | Add HuggingFace Hub usage example | | `docs/features/` | Document Hub integration | | `tests/test_hub.py` | Add mixin tests | ## Lint Scope - Run `pytest tests/test_hub*.py -v` - Run `mypy nightmarenet/hub/` - Verify `huggingface-hub` version compatible with `PyTorchModelHubMixin` - Test: `from nightmarenet.hub import NightmareNetModelHubMixin` works > **Important:** Please read CONTRIBUTING.md in full before requesting assignment. Assignment requests without a technical approach will not be considered. PRs demonstrating thorough research, benchmarks, or creative approaches beyond minimum criteria may earn bonus XP labels.