feat: train_hyper.py v3 — full architecture, optimized forward + MeZO, no features cut

fix: train_hyper.py v2 — lean mode, reduced layers, no overhead, 10k+ tok/s target

feat: config.json v5.3 — add hyper_training section with 7 paradigms

docs: update README for v5.3 — document 7 HYPER training paradigms

feat: register hyper paradigms in chimera/__init__.py (v5.3)

feat: add train_hyper.py — 7-paradigm stacked training for 10k+ tok/s on CPU

feat: add chimera/hyper.py — 7 paradigms engine for 10k+ tok/s CPU training

Upload train_fast.py

Upload inference.py

Upload inference.py

Upload chimera/model.py with huggingface_hub

Upload chimera/evolution.py with huggingface_hub

Upload chimera/__init__.py with huggingface_hub

Upload folder using huggingface_hub

initial commit