fix: log both stdout+stderr on push fail, rate-limit backoff, pure pip install no cmake

fix: force --only-binary for llama-cpp-python on Kaggle Python 3.12

fix: force --only-binary for llama-cpp-python on Kaggle Python 3.12

fix: kaggle sm_75 + pre-built wheel, OpenClaw+Ollama on clean base, no health-check

fix: replace openclaw health-check image with python:3.11-slim, eliminate spurious pauses

fix: move ollama install to start.sh for faster build, prevent Space pause during build

fix: dataset_sources attached, kernel errors non-fatal, space stays running

fix: remove random suffix from Kaggle kernel slug to match actual stored name

fix: kaggle is_private bool False for kagglesdk v0.1.22

fix: python status server on 8080 so nginx never 502s regardless of openclaw

fix: python json config (no heredoc), fallback http server on 8080

fix: sync_hub clean rewrite, openclaw password key, 7b model, sleep infinity keepalive

fix: kaggle 60s poll delay, is_private string, openclaw password, sync_hub

fix: sync_hub syntax, openclaw password key, is_private bool, remove double logging, 14b model

fix: add zstd for ollama install

fix: add USER root before apt-get in Dockerfile

fix: remove secrets and token from repo, add gitignore

clean: full pipeline, no secrets in history