HeatOps v2: multi-crew scheduling and reproducible evaluation

[MichelleHan7]

Why

The hackathon scheduler selects only one crew and depends on a small historical demo. This branch adds workforce assignment and explicit experimental evidence while preserving the Phoenix regression.

Changes

• Multi-crew CP-SAT engine with skills, shifts, availability, delay budgets and independent feasibility checks.
• Optional consecutive-task travel constraints and estimated route metrics.
• Seeded synthetic scenarios, four-policy benchmarks, failure-inclusive reporting and model diagnostics.
• Provenance-aware Open-Meteo/FortyGuard/offline providers, fleet Streamlit workspace, custom input, CLI and exports.

Validation

162 local tests passed; Ruff lint/format and compileall passed. The original Phoenix result remains 2.971485%, with an unchanged snapshot checksum. Streamlit HTTP health and a real Open-Meteo request were verified.

The complete fixed benchmark rerun and final documentation are being attached before handoff. This PR remains a draft. No main-branch merge or public deployment change is requested.