Hygiene, KPI boundary tests and mutation testing

[vidit-16]

Small hygiene pass. The repo already had CI, Docker, docs, a license and a large test suite, so this only fills gaps.

Changes

• Hygiene: removed a committed local tool config file (.claude/launch.json). .gitignore now covers .env*, coverage output, mutation output and IDE folders. Added .env.example for the optional OPENAI_API_KEY.
• Mutation testing: added scripts/mutation_test.py, an AST-based mutation tester (mutmut doesn't run on Windows). It swaps comparison and arithmetic operators and nudges integer constants, one at a time, and restores the file afterwards.
• Tests: added tests/test_kpis_edges.py with 8 boundary tests: zero and unit denominators for fill rate, cancellation rate, mix, growth and target variance, a missing sales type, and rounding precision.
• README: updated the test count and added a mutation testing row.

Results

• ruff check .: passes
• pytest: 316 passed (308 existing + 8 new) locally on Windows
• KPI mutation score: 13/37 (35%) → 34/37 (92%). The 3 survivors are equivalent mutants (> 0 → >= 0 where the count is never 0, or where 0/0 is already NaN).

Not changed

• Dockerfile, CI and docs were already in good shape.
• No ruff format: it would reformat 50 files for no functional gain.
• No new A/B comparison: scripts/evaluate_forecasts.py already benchmarks five forecasting models.