feat: Phase 2 — recommender engine with 15+ test scenarios

[bakodramane]

Pure function recommend(DataAvailability) → Recommendation[] that maps user data flags to a ranked list of SAE methods with rationale and caveats.

Implements §3 priority table from docs/SAE-CATALOGUE.md:

• Filters methods by required inputs and target type
• Scores methods by scenario (poverty, binary/proportion, count, continuous)
• Handles outlier, spatial, out-of-sample, and weights sub-scenarios
• Always appends direct estimator last as a benchmark
• Sets stataV14Warning for methods with stataMinVersion > 14
• Builds whyApplicable strings and inferred caveats per scenario

380 tests pass (56 new recommender tests + 324 existing).