Add Churn Explorer Streamlit app, Docker image and CI

[Martinoor]

Summary

• src/churn_app: an installable package for loading event logs (parquet/CSV from a path, bytes or upload; batched reads with date-range and deterministic user sampling), schema validation, PII dropping, filtering (EventFilter), churn labels (ever / competition horizon), leakage-safe user features, and sklearn models with the threshold tuned for balanced accuracy.
• app/streamlit_app.py: Overview, Segments, User drill-down, Churn model and Data tabs. Runs on a deterministic synthetic sample, or on the real Kaggle files when they are mounted at /data or uploaded.
• Tests: 103 tests, 98% branch coverage. They include detailed tests of the import and filtering functions, plus headless AppTest end-to-end runs.
• Container: multi-stage Dockerfile with digest-pinned bases, uv.lock --frozen, a non-root user and a healthcheck. docker-compose.yml is included.
• CI: lockfile check, ruff, pytest on 3.12/3.13, then image build and container smoke test. On main and tags it pushes a multi-arch image to ghcr.io/martinoor/churn-explorer.
• Dependabot, pre-commit, Makefile, README docs. The research code is untouched.

Test plan

• make lint and make test pass locally (103 passed)
• docker build works; the container turns healthy, runs as UID 10001 and reads a parquet file mounted at /data
• Every tab checked in the browser, including model training
• CI green on this PR

🤖 Generated with Claude Code