Add a multi-map replay explorer

[Revincxt]

Summary

• replace the single-map, presentation-heavy demo with a restrained research replay explorer
• add four structurally distinct 16×12 warehouse cases and a versioned export-gallery pipeline
• add synchronized A/B controller inspection, step/speed controls, event seeking, a bounded future-path preview, and explicit trace-completion states
• record the root seed and scenario fingerprints, and represent unmeasured decision timing as null
• make replay-buffer sampling copy only the selected batch instead of the whole buffer
• strengthen GitHub Pages and root-path deployment checks

Maps

• Rack maze
• Parallel aisles
• Cross-dock
• Serpentine

Each case is a real scenario with its own obstacle topology, charging positions, dynamic closures, trained agents, and committed simulator traces.

Verification

• ruff check .
• mypy src
• pytest --cov=adaptive_agent_lab --cov-report=term-missing (178 passed, 81 subtests, 85.14% coverage)
• pnpm run lint
• pnpm run test:sites
• pnpm run test:pages
• git diff --check