release: ship the MLZOO advanced ML model zoo (A–C2) to main

[w7-mgfcode]

Tracking issue for the dev → main release PR.

main is at v0.2.15; dev has accumulated the full MLZOO advanced-forecasting arc plus fixes. Merging the release PR lets release-please open its Release PR and tag the next version (pre-1.0 feat: → PATCH, so v0.2.16).

In this release

• MLZOO-A (Feature-aware forecasting foundation (MLZOO-A / PRP-29) #238) — feature-aware forecasting foundation, shared feature-frame contract
• MLZOO-B (feat(forecast): add LightGBM feature-aware forecasting model (MLZOO-B) #242) — LightGBM feature-aware model (ml-lightgbm extra)
• MLZOO-B.2 (feat(backtest): wire feature-aware models into backtesting (MLZOO-B.2) #244) — feature-aware models wired into the backtesting fold loop
• MLZOO-C1 (feat(forecast): add XGBoost feature-aware forecasting model (#247) #251) — XGBoost feature-aware model (ml-xgboost extra)
• MLZOO-C2 (feat(forecast): add Prophet-like additive forecasting model (#248) #250) — Prophet-like additive model (pure scikit-learn)
• fix(api) (fix(api): allow Tailscale CGNAT origins in dev CORS allow-list (#246) #249) — Tailscale CGNAT origins in dev CORS allow-list
• fix(ui/jobs) (fix(ui): What-If Planner submits empty assumption dates → backend 422 #228, fix(ui): model_exogenous scenario path is unreachable from the What-If Planner #229) — what-if planner empty-date + model_exogenous reachability
• docs + chore (PRPs, MLZOO briefs, uv.lock sync)

Closed via the dev → main release PR.